alibabaresearch / advancedliteratemachinery

A collection of original, innovative ideas and algorithms towards Advanced Literate Machinery. This project is maintained by the OCR Team in the Language Technology Lab, Tongyi Lab, Alibaba Group.

Advanced Literate Machinery

Introduction

The ultimate goal of our research is to build a system that has high-level intelligence, i.e., possessing the abilities to read, think and create, so advanced that it could even surpass human intelligence one day in the future. We name this kind of systems Advanced Literate Machinery (ALM).

To start with, we currently focus on teaching machines to read from images and documents. In years to come, we will explore the possibilities of endowing machines with the intellectual capabilities of thinking and creating, catching up with and surpassing GPT-4 and GPT-4V.

This project is maintained by the 读光 OCR Team (读光-Du Guang means “Reading The Light”) in the Tongyi Lab, Alibaba Group.

Visit our 读光-Du Guang Portal and DocMaster to experience online demos for OCR and Document Understanding.

Recent Updates

2024.4 Release

• OmniParser (OmniParser: A Unified Framework for Text Spotting, Key Information Extraction and Table Recognition, CVPR 2024. paper): We propose a universal model for parsing visually-situated text across diverse scenarios, called OmniParser, which can simultaneously handle three typical visually-situated text parsing tasks: text spotting, key information extraction, and table recognition. In OmniParser, all tasks share the unified encoder-decoder architecture, the unified objective: point-conditioned text generation, and the unified input & output representation: prompt & structured sequences.

2024.3 Release

• GEM (GEM: Gestalt Enhanced Markup Language Model for Web Understanding via Render Tree, EMNLP 2023. paper): Web pages serve as crucial carriers for humans to acquire and perceive information. Inspired by the Gestalt psychological theory, we propose an innovative Gestalt Enhanced Markup Language Model (GEM for short) for hosting heterogeneous visual information from render trees of web pages, leading to excellent performances on tasks such as web question answering and web information extraction.

2023.9 Release

• DocXChain (DocXChain: A Powerful Open-Source Toolchain for Document Parsing and Beyond, arXiv 2023. report): To promote the level of digitization and structurization for documents, we develop and release an open-source toolchain, called DocXChain, for precise and detailed document parsing. Currently, basic capabilities, including text detection, text recognition, table structure recognition, and layout analysis, are provided. Also, typical pipelines, i.e., general text reading, table parsing, and document structurization, are built to support more complicated applications related to documents. Most of the algorithmic models are from ModelScope. Formula recognition (using models from RapidLatexOCR) and whole PDF conversion (PDF to JSON format) are now supported.
• LISTER (LISTER: Neighbor Decoding for Length-Insensitive Scene Text Recognition, ICCV 2023. paper): We propose a method called Length-Insensitive Scene TExt Recognizer (LISTER), which remedies the limitation regarding the robustness to various text lengths. Specifically, a Neighbor Decoder is proposed to obtain accurate character attention maps with the assistance of a novel neighbor matrix regardless of the text lengths. Besides, a Feature Enhancement Module is devised to model the long-range dependency with low computation cost, which is able to perform iterations with the neighbor decoder to enhance the feature map progressively..
• VGT (Vision Grid Transformer for Document Layout Analysis, ICCV 2023. paper): To fully leverage multi-modal information and exploit pre-training techniques to learn better representation for document layout analysis (DLA), we present VGT, a two-stream Vision Grid Transformer, in which Grid Transformer (GiT) is proposed and pre-trained for 2D token-level and segment-level semantic understanding. In addition, a new benchmark for assessing document layout analysis algorithms, called D^4LA, is curated and released.
• VLPT-STD (Vision-Language Pre-Training for Boosting Scene Text Detectors, CVPR 2022. paper): We adapt vision-language joint learning for scene text detection, a task that intrinsically involves cross-modal interaction between the two modalities: vision and language. The pre-trained model is able to produce more informative representations with richer semantics, which could readily benefit existing scene text detectors (such as EAST and DB) in the down-stream text detection task.

2023.6 Release

• LiteWeightOCR (Building A Mobile Text Recognizer via Truncated SVD-based Knowledge Distillation-Guided NAS, BMVC 2023. paper): To make OCR models deployable on mobile devices while keeping high accuracy, we propose a light-weight text recognizer that integrates Truncated Singular Value Decomposition (TSVD)-based Knowledge Distillation (KD) into the Neural Architecture Search (NAS) process.

2023.4 Release

• GeoLayoutLM (GeoLayoutLM: Geometric Pre-training for Visual Information Extraction, CVPR 2023. paper): We propose a multi-modal framework, named GeoLayoutLM, for Visual Information Extraction (VIE). In contrast to previous methods for document pre-training, which usually learn geometric representation in an implicit way, GeoLayoutLM explicitly models the geometric relations of entities in documents.

2023.2 Release

• LORE-TSR (LORE: Logical Location Regression Network for Table Structure Recognition, AAAI 2022. paper): We model Table Structure Recognition (TSR) as a logical location regression problem and propose a new algorithm called LORE, standing for LOgical location REgression network, which for the first time combines logical location regression together with spatial location regression of table cells.

2022.9 Release

• MGP-STR (Multi-Granularity Prediction for Scene Text Recognition, ECCV 2022. paper): Based on ViT and a tailored Adaptive Addressing and Aggregation module, we explore an implicit way for incorporating linguistic knowledge by introducing subword representations to facilitate multi-granularity prediction and fusion in scene text recognition.
• LevOCR (Levenshtein OCR, ECCV 2022. paper): Inspired by Levenshtein Transformer, we cast the problem of scene text recognition as an iterative sequence refinement process, which allows for parallel decoding, dynamic length change and good interpretability.

advancedliteratemachinery's Issues

LevOCR: how to train the LevOCR?

is there a train.py for LevOCR training?

cuda11以上是否可以？

您好， 我看您说用DLA backbone，推荐cuda10.1，请问cuda11以上版本是否可以？

LORE的loss部分

您好，非常抱歉打扰到您，有两个问题想向您请教一下。
1、在代码lore中没有发现论文中所提到的inner，inter loss函数。请问是我没找到，还是代码中没有公开？
2、在您公开的wtw数据集label中和下载的wtw中的lable和图像不能完全对应，能否提供你们做实验的那版本数据集。非常感谢。

D4LA Dataset

Hello, when will the D4LA Dataset be released? This is a great dataset and I can't wait to test the effect.

about execution of MGP-STR code

I encountered this error while executing MGP-STR code：

ValueError: Connection error, and we cannot find the requested files in the cached path. Please try again or make sure your Internet connection is on.

May I ask how to solve it? Thank you!

pairing of key value

I want to know how to proceed with key-value pairing only because I have label previously.

LORE相关预训练模型

请问。LORE相关的预训练模型何时能放出来？如训练过程用到的model_best.pth和processor_best.pth。

Loss收敛问题

  请教一下 LORE 训练自己的数据集时，hm_loss, wh_loss, ax_loss, sax_loss 分别收敛到多少会获得一个比较好的解析效果？另外对于非readme提到的数据集，除了max_objs,max_pairs与max_cors这三个参数，还有什么参数需要修改的？
   非常期待您的回复！

How to pre-training?

Hi. Thanks for the great research.

I would like to train GeoLayoutLM using a dataset consisting of other languages, is it possible to provide a pre-trainable code? Or could you give me a tip, for pre-training, I think I need the loss configurations for these 3: GeoPair, GeoMPair and GeoTriplet

Increase max_seq_length

How can we pass more tokens than max_seq_length.
I have more tokens than 512

Paper link of LORE-TSR

Hi Authors, Could you please provide the paper dpf link of LORE-TSR ? Thanks !

model_ckpt issue

Hi! Thank you very much for open-sourcing the project code!
But I encountered a problem when running train.py FileNotFoundError: [Errno 2] No such file or directory: 'path/to/geolayoutlm_large_pretrain.pt', and ended up in configs/ GeoLayoutLM/finetune_funsd.yaml to find the source model_ckpt: path/to/geolayoutlm_large_pretrain.pt.
May I ask if this model path is not open source in the project list? Or maybe it is open source but unfortunately I didn't find it? Hopefully this can be resolved quickly! Thanks a lot!
The full error message is attached below:
2023-09-12 09:40:39,088 geolayoutlm_vie.py:60 [INFO] init weight from path/to/geolayoutlm_large_pretrain.pt'
Traceback (most recent call last):
File "train.py", line 55, in <module>
main()
File "train.py", line 44, in main
pl_module = GeoLayoutLMVIEModule(cfg)
File "/geoLayoutlm/lightning_modules/geolayoutlm_vie_module.py", line 20, in __init__
super().__init__(cfg)
File "/geoLayoutlm/lightning_modules/bros_module.py", line 24, in __init__
self.net = get_model(self.cfg)
File "/geoLayoutlm/model/__init__.py", line 5, in get_model
model = GeoLayoutLMVIEModel(cfg=cfg)
File "/geoLayoutlm/model/geolayoutlm_vie.py", line 56, in __init__
self._init_weight()
File "/geoLayoutlm/model/geolayoutlm_vie.py", line 61, in _init_weight
state_dict = torch.load(model_path, map_location='cpu')
File "/opt/conda/lib/python3.7/site-packages/torch/serialization.py", line 579, in load
with _open_file_like(f, 'rb') as opened_file:
File "/opt/conda/lib/python3.7/site-packages/torch/serialization.py", line 230, in _open_file_like
return _open_file(name_or_buffer, mode)
File "/opt/conda/lib/python3.7/site-packages/torch/serialization.py", line 211, in __init__
super(_open_file, self).__init__(open(name, mode))
FileNotFoundError: [Errno 2] No such file or directory: 'path/to/geolayoutlm_large_pretrain.pt'

Entity Grouping function for GeoLayoutLM

Hello team!

Thanks for such a great contribution to the open-source community.

Can you please explain about the logic behind entity-grouping done to identify the entities before linking key-value pairs. It would be great if you can also point me to that respective function in the repo.

Thanks!

关于部署geoLayoutLM

非常感谢你们的工作，我们在自己数据集上试了一下效果不错，现在准备部署，按照传统的方式我们准备转onnx，但是遇到问题，无法转onnx。
请问下对该模型的部署有什么建议吗？
非常感谢！

missing requirements.txt file

I can't find the requirements.txt file in the root folder.

///

Can't pickle local object 'linear_scheduler.<locals>.lr_lambda'

when i use multiple gpus to train GeoLayoutLM, error as follows:

Can't pickle local object 'linear_scheduler..lr_lambda'

@wdp-007 @congyao @alibaba-oss Looking forward to your reply.

LISTER

Will the "LISTER: Neighbor Decoding for Length-Insensitive Scene Text Recognition" be introduced?

请问Accuracy指标中的AVG是怎么算的呢？在线等，谢谢

关于geolayoutLM的模型输入

你好，
请问下模型输入中：bbox_4p_normalized，line_rank_id，line_rank_inner_id是什么意思？

geolayout的modelscope上的中文模型

作者您好，感谢作者release了这么好的工作，其中relation heads在提点方面有很大的作用，在modelscope上多语言的模型中没有relation heads部分的参数，如果不load relation heads直接训funsd，大概只能到70%+，relation heads部分的参数方便提供么？

Inference GeoLayoutLM

Hello! Thanks a lot for the great work!
I have a question regarding model inference. As far as I can see, the model forward requires block boxes to be fed. But in order to get the block boxes, you must first get them, for example, using LayoutLM.
Is it possible to run an inference of model with only an image, word boxes, and their corresponding text as input? Could you provide the code for the inference?

请教一下，LORE的backbone改成resnet50，使用PubTabNet-Dataset收敛效果不太好？

配置：

• 训练的配置文件：
  

• src/lib/models/networks/fpn_mask_resnet.py:class PoseResNet内容为：
  

loss：

看loss，感觉收敛的效果并不是特别好，尤其是ax_l的值不怎么下降，请问这是什么原因导致的？

LORE's labels

Could you please tell me what's the meaning of the labels of LORE, especially the segmentation, logic_axis, area, iscrowd and ignore in annotations

PubTabNet获取逻辑坐标和无线表格问题

大佬你好，我有两个问题：
1.请问PubTabNet如何获取逻辑坐标呀？
2.对于无线表格，是不是无法预测空cell的逻辑坐标？后处理的话，是不是无法判断两个或多个相邻空cell到底是几个cell？

如果要识别多类，把preprocess.py中classes改成实际标记labels，assert

RuntimeError: CUDA error: device-side assert triggered
Compile with TORCH_USE_CUDA_DSA to enable device-side assertions

改成3类label，能跑起来，实际337类就不行，有哪些地方需要特别注意么

LevOCR on custom datasets

Can i finetune LevOCR on my custom dataset for my language (Vietnamese) ?

提供的coco数据中logic_axis和wtw原始数据标注中startcol、endcol不一致

logic_axis是怎么转换的呢，和wtw原始的xml标注对比了一下，差别很大

MGP-STR在中文效果如何？

不同大小模型的初始学习率是否相同？

Hi，请问除了base模型以外的其他模型是否需要设置不同的学习率？如果是的话，推荐是多少呢？

Visualize the GeoLayoutLM results?

Here I used the pre-trained model from ModelScope but I don't know how to get the head outputs and visualize the text and image embedding for the respective image. Here I attached the ModelScope code for reference and link also.

from modelscope.models import Model
from modelscope.pipelines import pipeline

model = Model.from_pretrained('damo/multi-modal_convnext-roberta-base_vldoc-embedding')
doc_VL_emb_pipeline = pipeline(task='document-vl-embedding', model=model)

inp = {
    'images': ['data/demo.png'], 
    'ocr_info_paths': ['data/demo.json']
}
result = doc_VL_emb_pipeline(inp)

print('Results of VLDoc: ')
for k, v in result.items():
    print(f'{k}: {v.size()}')
# Expected output:
# img_embedding: torch.Size([1, 151, 768]), 151 = 1 global img feature + 150 segment features
# text_embedding: torch.Size([1, 512, 768])

Thanks in advance!

demo_imgs.py not working

Thank you for sharing your incredible work. I was trying to run the levocr pretrained model on couple of demo images.
demo_imgs script wasn't running, resolved it by making few changes in the demo_imgs.py

Following are the changes I have made -

At Line 141 model.module.load_state_dict(torch.load(opt.saved_model, map_location=device),strict=False)
At Line 150 AlignCollate_demo = AlignCollateTest(imgH=opt.imgH, imgW=opt.imgW)
At Line 171 vision_final_pred, _ = converter.encode_levt(vision_preds_str, src_dict, device=device, batch_max_length=pred_vision.size(1))

So,when I ran the script, this was the output

Can you help me to interpret the output or the right way to run the demo script?

MGP-STR：where is the impl of self.cls_token?

"cls_tokens = self.cls_token.expand(B, -1, -1) # stole cls_tokens impl from Phil Wang, thanks", I can't find the impl of self.cls_token

LORE-TSR's TEDS metrics?

How does the LORE-TSR calculate TEDS indicators on the PubTabNet dataset?
How is the blank cell situation handled?

Regarding the inference relation head in the GeoLayoutLM and LayoutLMv3 models

I would like to express my gratitude for your hard work.
Upon reviewing the benchmark table, I have observed the impressive accuracy of the relation head in GeoLayoutLM. Would it be possible for you to share the code for pre-produce the result of LayoutLMv3 or provide guidance on how to integrate it with the current source code?

LORE-TSR使用标注格式

您好，我想测试LORE-TSR在PubTabNet，SciTSR和WTW数据集上的性能，但这些数据集提供的标注格式都不同。PubTabNet为jsonl，SciTSR为chunk的structure，WTW为xml。请问我是否需要给出统一的标注类型，且标注的格式为？

LevOCR字符集问题

你好，如果我对于字符集的检测除了大小写字母和数字以外还需要一些特殊符号例如-|~等，是不是需要重新训练整个模型，如果需要的话请问8卡A100训练ST+MJ数据集预估需要多久

关于GeoLayoutLM中文支持模型

非常感谢达摩院读光组的工作，GeoLayoutLM是个很棒的模型，但是这个模型使用的bert base的tokenizer，请问有预训练对中文支持的模型吗，或者未来会发布吗？
@wdp-007 @alibaba-oss @congyao
非常感谢！

How to deploy the GeoLayoutLM model to modelscopes or huggingface?

Now we have no reference function, please help me publish the model

Change bert

Hello!!

I want to learn using Korean dataset. Can I change the bert model?? If so, which part should I modify?

LORE finetuning gives wierd bounding boxes.

Hello, thank you for the great work you did with LORE.

I am trying finetuning the small version of LORE on a custom dataset but it seems that when I do the model results are getting worst.
I tried some zero-shot inference with ckpt_ptn model weights.
For finetuning I used

!python main.py ctdet_small \
	--dataset table_small \
	--exp_id CTU_fold_0 \
	--dataset_name ctu_fold_0 \
	--image_dir /datadrive/images/CTU \
	--wiz_2dpe \
	--wiz_stacking \
	--tsfm_layers 3 \
	--stacking_layers 3 \
	--batch_size 16 \
	--master_batch 12 \
	--arch dla_34 \
	--lr 1e-4 \
	--K 500 \
	--MK 1000 \
	--num_epochs 10 \
	--lr_step '100, 160' \
	--gpus 0 \
	--num_workers 16 \
	--val_intervals 1 \
	--save_all \
	--load_model /datadrive/LORE_experiments/pretrined_checkpoints/small_model_best.pth \
	--load_processor /datadrive/LORE_experiments/pretrined_checkpoints/small_processor_best.pth

where small_model_best.pth and small_processor_best.pth are the names I used to save the ckpt_ptn model weights.

I also produced annotations similar to the ones you provided in the subset of PubTabNet. I can provide an example if necessary.

Am I doing something wrong in the script?
Thank in advance!

MJ和ST数据集图片是灰白的吗？image的channel为1啊。

Geolayoutlm train.py issue

Hello team!

Thanks for such a great contribution to the open-source community.

But sadly I am stuck at one thing, at the end of train.py, it is thowing AttributeError: 'VIEDataModule' object has no attribute '_has_setup_TrainerFn.FITTING'. How to resolve this issue. Kindly help me.

Code Snippet
if cfg.model.head in ["vie"]:
pl_module = GeoLayoutLMVIEModule(cfg)
else:
raise ValueError(f"Not supported head {cfg.model.head}")

# import ipdb;ipdb.set_trace()
data_module = VIEDataModule(cfg, pl_module.net.tokenizer)

trainer.fit(model = pl_module, datamodule=data_module)

Error Snippet
Exception has occurred: AttributeError
'VIEDataModule' object has no attribute '_has_setup_TrainerFn.FITTING'
File "D:\TEST\GeoLayoutLM\train.py", line 52, in main
trainer.fit(model = pl_module, datamodule=data_module)
File "D:\TEST\GeoLayoutLM\train.py", line 56, in
main()
AttributeError: 'VIEDataModule' object has no attribute '_has_setup_TrainerFn.FITTING'

MagickWand错误，按照提示，不能下载

libraries = load_library()出错

ImportError: MagickWand shared library not found.
You probably had not installed ImageMagick library.

Try to install:
apt-get install libmagickwand-dev

请问有遇到过这种问题吗？已经按照说明，也尝试了各种源，但是不能下载

About NMS merge output to avoid Overlap Cells?

Hi, LORE model is great. I have a question, i saw the command in scripts infer have argument --nms but in file opts.py thís argument have action 'store_false'. It means we don't use nms function. May i ask why? Cause i run some sample images, it gave me lots of overlap cell. I think using nms can reduce this problem.

preprocess for CORD datasets

How did you preprocessed in GeoLayoutLM regarding CORD datasets?

train.py error

I cannot run train.py.
`python train.py --config=configs/finetune_funsd.yaml

Traceback (most recent call last):
File "train.py", line 6, in
from lightning_modules.data_modules.vie_data_module import VIEDataModule
File "/home/me/projects/UzPostOCR/AdvancedLiterateMachinery/DocumentUnderstanding/GeoLayoutLM/lightning_modules/data_modules/vie_data_module.py", line 8, in
import cv2
File "/home/me/miniconda3/envs/geo/lib/python3.8/site-packages/cv2/init.py", line 181, in
bootstrap()
File "/home/me/miniconda3/envs/geo/lib/python3.8/site-packages/cv2/init.py", line 175, in bootstrap
if __load_extra_py_code_for_module("cv2", submodule, DEBUG):
File "/home/me/miniconda3/envs/geo/lib/python3.8/site-packages/cv2/init.py", line 28, in __load_extra_py_code_for_module
py_module = importlib.import_module(module_name)
File "/home/me/miniconda3/envs/geo/lib/python3.8/importlib/init.py", line 127, in import_module
return _bootstrap._gcd_import(name[level:], package, level)
File "/home/me/miniconda3/envs/geo/lib/python3.8/site-packages/cv2/gapi/init.py", line 301, in
cv.gapi.wip.GStreamerPipeline = cv.gapi_wip_gst_GStreamerPipeline
AttributeError: partially initialized module 'cv2' has no attribute 'gapi_wip_gst_GStreamerPipeline' (most likely due to a circular import)`

LORE模型demo_wireless.sh运行

在运行LORE模型的demo_wireless.sh时，运行结果出错。
demo_wireless.sh
python demo.py ctdet_mid \ --dataset table_mid \ --demo ../input_images/wireless \ --demo_name demo_wireless \ --debug 1 \ --arch resfpnhalf_18 \ --K 3000 \ --MK 5000 \ --upper_left \ --tsfm_layers 4\ --stacking_layers 4 \ --gpus 0\ --wiz_2dpe \ --wiz_detect \ --wiz_stacking \ --convert_onnx 0 \ --vis_thresh_corner 0.3 \ --vis_thresh 0.2 \ --scores_thresh 0.2 \ --nms \ --demo_dir ../output/test_wireless/ \ --load_model ../dir_of_ckpt/ckpt_wireless/model_best.pth \ --load_processor ../dir_of_ckpt/ckpt_wireless/processor_best.pth
预训练模型能够正常加载，但是存在一些问题，
Fix size testing. training chunk_sizes: [1] The output will be saved to /root/data1/bwq/LORE_new/AdvancedLiterateMachinery/DocumentUnderstanding/LORE-TSR/src/lib/../../exp/ctdet_mid/default heads {'hm': 2, 'st': 8, 'wh': 8, 'ax': 256, 'cr': 256, 'reg': 2} loaded ../dir_of_ckpt/ckpt_wireless/model_best.pth, epoch 100 Drop parameter base.base_layer.0.weight. Drop parameter base.base_layer.1.weight. Drop parameter base.base_layer.1.bias. Drop parameter base.base_layer.1.running_mean. Drop parameter base.base_layer.1.running_var. Drop parameter base.base_layer.1.num_batches_tracked. Drop parameter base.level0.0.weight. Drop parameter base.level0.1.weight. Drop parameter base.level0.1.bias. Drop parameter base.level0.1.running_mean. Drop parameter base.level0.1.running_var. Drop parameter base.level0.1.num_batches_tracked. Drop parameter base.level1.0.weight. Drop parameter base.level1.1.weight. Drop parameter base.level1.1.bias. Drop parameter base.level1.1.running_mean. Drop parameter base.level1.1.running_var. Drop parameter base.level1.1.num_batches_tracked. Drop parameter base.level2.tree1.conv1.weight. Drop parameter base.level2.tree1.bn1.weight. Drop parameter base.level2.tree1.bn1.bias. Drop parameter base.level2.tree1.bn1.running_mean. Drop parameter base.level2.tree1.bn1.running_var. Drop parameter base.level2.tree1.bn1.num_batches_tracked. Drop parameter base.level2.tree1.conv2.weight. Drop parameter base.level2.tree1.bn2.weight. Drop parameter base.level2.tree1.bn2.bias. Drop parameter base.level2.tree1.bn2.running_mean. Drop parameter base.level2.tree1.bn2.running_var. Drop parameter base.level2.tree1.bn2.num_batches_tracked. Drop parameter base.level2.tree2.conv1.weight. Drop parameter base.level2.tree2.bn1.weight. Drop parameter base.level2.tree2.bn1.bias. Drop parameter base.level2.tree2.bn1.running_mean. Drop parameter base.level2.tree2.bn1.running_var. Drop parameter base.level2.tree2.bn1.num_batches_tracked. Drop parameter base.level2.tree2.conv2.weight. Drop parameter base.level2.tree2.bn2.weight. Drop parameter base.level2.tree2.bn2.bias. Drop parameter base.level2.tree2.bn2.running_mean. Drop parameter base.level2.tree2.bn2.running_var. Drop parameter base.level2.tree2.bn2.num_batches_tracked. Drop parameter base.level2.root.conv.weight. Drop parameter base.level2.root.bn.weight. Drop parameter base.level2.root.bn.bias. Drop parameter base.level2.root.bn.running_mean. Drop parameter base.level2.root.bn.running_var. Drop parameter base.level2.root.bn.num_batches_tracked. Drop parameter base.level2.project.0.weight. Drop parameter base.level2.project.1.weight. Drop parameter base.level2.project.1.bias. Drop parameter base.level2.project.1.running_mean. Drop parameter base.level2.project.1.running_var. Drop parameter base.level2.project.1.num_batches_tracked. Drop parameter base.level3.tree1.tree1.conv1.weight. Drop parameter base.level3.tree1.tree1.bn1.weight. Drop parameter base.level3.tree1.tree1.bn1.bias. Drop parameter base.level3.tree1.tree1.bn1.running_mean. Drop parameter base.level3.tree1.tree1.bn1.running_var. Drop parameter base.level3.tree1.tree1.bn1.num_batches_tracked. Drop parameter base.level3.tree1.tree1.conv2.weight. Drop parameter base.level3.tree1.tree1.bn2.weight. Drop parameter base.level3.tree1.tree1.bn2.bias. Drop parameter base.level3.tree1.tree1.bn2.running_mean. Drop parameter base.level3.tree1.tree1.bn2.running_var. Drop parameter base.level3.tree1.tree1.bn2.num_batches_tracked. Drop parameter base.level3.tree1.tree2.conv1.weight. Drop parameter base.level3.tree1.tree2.bn1.weight. Drop parameter base.level3.tree1.tree2.bn1.bias. Drop parameter base.level3.tree1.tree2.bn1.running_mean. Drop parameter base.level3.tree1.tree2.bn1.running_var. Drop parameter base.level3.tree1.tree2.bn1.num_batches_tracked. Drop parameter base.level3.tree1.tree2.conv2.weight. Drop parameter base.level3.tree1.tree2.bn2.weight. Drop parameter base.level3.tree1.tree2.bn2.bias. Drop parameter base.level3.tree1.tree2.bn2.running_mean. Drop parameter base.level3.tree1.tree2.bn2.running_var. Drop parameter base.level3.tree1.tree2.bn2.num_batches_tracked. Drop parameter base.level3.tree1.root.conv.weight. Drop parameter base.level3.tree1.root.bn.weight. Drop parameter base.level3.tree1.root.bn.bias. Drop parameter base.level3.tree1.root.bn.running_mean. Drop parameter base.level3.tree1.root.bn.running_var. Drop parameter base.level3.tree1.root.bn.num_batches_tracked. Drop parameter base.level3.tree1.project.0.weight. Drop parameter base.level3.tree1.project.1.weight. Drop parameter base.level3.tree1.project.1.bias. Drop parameter base.level3.tree1.project.1.running_mean. Drop parameter base.level3.tree1.project.1.running_var. Drop parameter base.level3.tree1.project.1.num_batches_tracked. Drop parameter base.level3.tree2.tree1.conv1.weight. Drop parameter base.level3.tree2.tree1.bn1.weight. Drop parameter base.level3.tree2.tree1.bn1.bias. Drop parameter base.level3.tree2.tree1.bn1.running_mean. Drop parameter base.level3.tree2.tree1.bn1.running_var. Drop parameter base.level3.tree2.tree1.bn1.num_batches_tracked. Drop parameter base.level3.tree2.tree1.conv2.weight. Drop parameter base.level3.tree2.tree1.bn2.weight. Drop parameter base.level3.tree2.tree1.bn2.bias. Drop parameter base.level3.tree2.tree1.bn2.running_mean. Drop parameter base.level3.tree2.tree1.bn2.running_var. Drop parameter base.level3.tree2.tree1.bn2.num_batches_tracked. Drop parameter base.level3.tree2.tree2.conv1.weight. Drop parameter base.level3.tree2.tree2.bn1.weight. Drop parameter base.level3.tree2.tree2.bn1.bias. Drop parameter base.level3.tree2.tree2.bn1.running_mean. Drop parameter base.level3.tree2.tree2.bn1.running_var. Drop parameter base.level3.tree2.tree2.bn1.num_batches_tracked. Drop parameter base.level3.tree2.tree2.conv2.weight. Drop parameter base.level3.tree2.tree2.bn2.weight. Drop parameter base.level3.tree2.tree2.bn2.bias. Drop parameter base.level3.tree2.tree2.bn2.running_mean. Drop parameter base.level3.tree2.tree2.bn2.running_var. Drop parameter base.level3.tree2.tree2.bn2.num_batches_tracked. Drop parameter base.level3.tree2.root.conv.weight. Drop parameter base.level3.tree2.root.bn.weight. Drop parameter base.level3.tree2.root.bn.bias. Drop parameter base.level3.tree2.root.bn.running_mean. Drop parameter base.level3.tree2.root.bn.running_var. Drop parameter base.level3.tree2.root.bn.num_batches_tracked. Drop parameter base.level3.project.0.weight. Drop parameter base.level3.project.1.weight. Drop parameter base.level3.project.1.bias. Drop parameter base.level3.project.1.running_mean. Drop parameter base.level3.project.1.running_var. Drop parameter base.level3.project.1.num_batches_tracked. Drop parameter base.level4.tree1.tree1.conv1.weight. Drop parameter base.level4.tree1.tree1.bn1.weight. Drop parameter base.level4.tree1.tree1.bn1.bias. Drop parameter base.level4.tree1.tree1.bn1.running_mean. Drop parameter base.level4.tree1.tree1.bn1.running_var. Drop parameter base.level4.tree1.tree1.bn1.num_batches_tracked. Drop parameter base.level4.tree1.tree1.conv2.weight. Drop parameter base.level4.tree1.tree1.bn2.weight. Drop parameter base.level4.tree1.tree1.bn2.bias. Drop parameter base.level4.tree1.tree1.bn2.running_mean. Drop parameter base.level4.tree1.tree1.bn2.running_var. Drop parameter base.level4.tree1.tree1.bn2.num_batches_tracked. Drop parameter base.level4.tree1.tree2.conv1.weight. Drop parameter base.level4.tree1.tree2.bn1.weight. Drop parameter base.level4.tree1.tree2.bn1.bias. Drop parameter base.level4.tree1.tree2.bn1.running_mean. Drop parameter base.level4.tree1.tree2.bn1.running_var. Drop parameter base.level4.tree1.tree2.bn1.num_batches_tracked. Drop parameter base.level4.tree1.tree2.conv2.weight. Drop parameter base.level4.tree1.tree2.bn2.weight. Drop parameter base.level4.tree1.tree2.bn2.bias. Drop parameter base.level4.tree1.tree2.bn2.running_mean. Drop parameter base.level4.tree1.tree2.bn2.running_var. Drop parameter base.level4.tree1.tree2.bn2.num_batches_tracked. Drop parameter base.level4.tree1.root.conv.weight. Drop parameter base.level4.tree1.root.bn.weight. Drop parameter base.level4.tree1.root.bn.bias. Drop parameter base.level4.tree1.root.bn.running_mean. Drop parameter base.level4.tree1.root.bn.running_var. Drop parameter base.level4.tree1.root.bn.num_batches_tracked. Drop parameter base.level4.tree1.project.0.weight. Drop parameter base.level4.tree1.project.1.weight. Drop parameter base.level4.tree1.project.1.bias. Drop parameter base.level4.tree1.project.1.running_mean. Drop parameter base.level4.tree1.project.1.running_var. Drop parameter base.level4.tree1.project.1.num_batches_tracked. Drop parameter base.level4.tree2.tree1.conv1.weight. Drop parameter base.level4.tree2.tree1.bn1.weight. Drop parameter base.level4.tree2.tree1.bn1.bias. Drop parameter base.level4.tree2.tree1.bn1.running_mean. Drop parameter base.level4.tree2.tree1.bn1.running_var. Drop parameter base.level4.tree2.tree1.bn1.num_batches_tracked. Drop parameter base.level4.tree2.tree1.conv2.weight. Drop parameter base.level4.tree2.tree1.bn2.weight. Drop parameter base.level4.tree2.tree1.bn2.bias. Drop parameter base.level4.tree2.tree1.bn2.running_mean. Drop parameter base.level4.tree2.tree1.bn2.running_var. Drop parameter base.level4.tree2.tree1.bn2.num_batches_tracked. Drop parameter base.level4.tree2.tree2.conv1.weight. Drop parameter base.level4.tree2.tree2.bn1.weight. Drop parameter base.level4.tree2.tree2.bn1.bias. Drop parameter base.level4.tree2.tree2.bn1.running_mean. Drop parameter base.level4.tree2.tree2.bn1.running_var. Drop parameter base.level4.tree2.tree2.bn1.num_batches_tracked. Drop parameter base.level4.tree2.tree2.conv2.weight. Drop parameter base.level4.tree2.tree2.bn2.weight. Drop parameter base.level4.tree2.tree2.bn2.bias. Drop parameter base.level4.tree2.tree2.bn2.running_mean. Drop parameter base.level4.tree2.tree2.bn2.running_var. Drop parameter base.level4.tree2.tree2.bn2.num_batches_tracked. Drop parameter base.level4.tree2.root.conv.weight. Drop parameter base.level4.tree2.root.bn.weight. Drop parameter base.level4.tree2.root.bn.bias. Drop parameter base.level4.tree2.root.bn.running_mean. Drop parameter base.level4.tree2.root.bn.running_var. Drop parameter base.level4.tree2.root.bn.num_batches_tracked. Drop parameter base.level4.project.0.weight. Drop parameter base.level4.project.1.weight. Drop parameter base.level4.project.1.bias. Drop parameter base.level4.project.1.running_mean. Drop parameter base.level4.project.1.running_var. Drop parameter base.level4.project.1.num_batches_tracked. Drop parameter base.level5.tree1.conv1.weight. Drop parameter base.level5.tree1.bn1.weight. Drop parameter base.level5.tree1.bn1.bias. Drop parameter base.level5.tree1.bn1.running_mean. Drop parameter base.level5.tree1.bn1.running_var. Drop parameter base.level5.tree1.bn1.num_batches_tracked. Drop parameter base.level5.tree1.conv2.weight. Drop parameter base.level5.tree1.bn2.weight. Drop parameter base.level5.tree1.bn2.bias. Drop parameter base.level5.tree1.bn2.running_mean. Drop parameter base.level5.tree1.bn2.running_var. Drop parameter base.level5.tree1.bn2.num_batches_tracked. Drop parameter base.level5.tree2.conv1.weight. Drop parameter base.level5.tree2.bn1.weight. Drop parameter base.level5.tree2.bn1.bias. Drop parameter base.level5.tree2.bn1.running_mean. Drop parameter base.level5.tree2.bn1.running_var. Drop parameter base.level5.tree2.bn1.num_batches_tracked. Drop parameter base.level5.tree2.conv2.weight. Drop parameter base.level5.tree2.bn2.weight. Drop parameter base.level5.tree2.bn2.bias. Drop parameter base.level5.tree2.bn2.running_mean. Drop parameter base.level5.tree2.bn2.running_var. Drop parameter base.level5.tree2.bn2.num_batches_tracked. Drop parameter base.level5.root.conv.weight. Drop parameter base.level5.root.bn.weight. Drop parameter base.level5.root.bn.bias. Drop parameter base.level5.root.bn.running_mean. Drop parameter base.level5.root.bn.running_var. Drop parameter base.level5.root.bn.num_batches_tracked. Drop parameter base.level5.project.0.weight. Drop parameter base.level5.project.1.weight. Drop parameter base.level5.project.1.bias. Drop parameter base.level5.project.1.running_mean. Drop parameter base.level5.project.1.running_var. Drop parameter base.level5.project.1.num_batches_tracked. Drop parameter base.fc.weight. Drop parameter base.fc.bias. Drop parameter dla_up.ida_0.proj_1.actf.0.weight. Drop parameter dla_up.ida_0.proj_1.actf.0.bias. Drop parameter dla_up.ida_0.proj_1.actf.0.running_mean. Drop parameter dla_up.ida_0.proj_1.actf.0.running_var. Drop parameter dla_up.ida_0.proj_1.actf.0.num_batches_tracked. Drop parameter dla_up.ida_0.proj_1.conv.weight. Drop parameter dla_up.ida_0.proj_1.conv.bias. Drop parameter dla_up.ida_0.proj_1.conv.conv_offset_mask.weight. Drop parameter dla_up.ida_0.proj_1.conv.conv_offset_mask.bias. Drop parameter dla_up.ida_0.up_1.weight. Drop parameter dla_up.ida_0.node_1.actf.0.weight. Drop parameter dla_up.ida_0.node_1.actf.0.bias. Drop parameter dla_up.ida_0.node_1.actf.0.running_mean. Drop parameter dla_up.ida_0.node_1.actf.0.running_var. Drop parameter dla_up.ida_0.node_1.actf.0.num_batches_tracked. Drop parameter dla_up.ida_0.node_1.conv.weight. Drop parameter dla_up.ida_0.node_1.conv.bias. Drop parameter dla_up.ida_0.node_1.conv.conv_offset_mask.weight. Drop parameter dla_up.ida_0.node_1.conv.conv_offset_mask.bias. Drop parameter dla_up.ida_1.proj_1.actf.0.weight. Drop parameter dla_up.ida_1.proj_1.actf.0.bias. Drop parameter dla_up.ida_1.proj_1.actf.0.running_mean. Drop parameter dla_up.ida_1.proj_1.actf.0.running_var. Drop parameter dla_up.ida_1.proj_1.actf.0.num_batches_tracked. Drop parameter dla_up.ida_1.proj_1.conv.weight. Drop parameter dla_up.ida_1.proj_1.conv.bias. Drop parameter dla_up.ida_1.proj_1.conv.conv_offset_mask.weight. Drop parameter dla_up.ida_1.proj_1.conv.conv_offset_mask.bias. Drop parameter dla_up.ida_1.up_1.weight. Drop parameter dla_up.ida_1.node_1.actf.0.weight. Drop parameter dla_up.ida_1.node_1.actf.0.bias. Drop parameter dla_up.ida_1.node_1.actf.0.running_mean. Drop parameter dla_up.ida_1.node_1.actf.0.running_var. Drop parameter dla_up.ida_1.node_1.actf.0.num_batches_tracked. Drop parameter dla_up.ida_1.node_1.conv.weight. Drop parameter dla_up.ida_1.node_1.conv.bias. Drop parameter dla_up.ida_1.node_1.conv.conv_offset_mask.weight. Drop parameter dla_up.ida_1.node_1.conv.conv_offset_mask.bias. Drop parameter dla_up.ida_1.proj_2.actf.0.weight. Drop parameter dla_up.ida_1.proj_2.actf.0.bias. Drop parameter dla_up.ida_1.proj_2.actf.0.running_mean. Drop parameter dla_up.ida_1.proj_2.actf.0.running_var. Drop parameter dla_up.ida_1.proj_2.actf.0.num_batches_tracked. Drop parameter dla_up.ida_1.proj_2.conv.weight. Drop parameter dla_up.ida_1.proj_2.conv.bias. Drop parameter dla_up.ida_1.proj_2.conv.conv_offset_mask.weight. Drop parameter dla_up.ida_1.proj_2.conv.conv_offset_mask.bias. Drop parameter dla_up.ida_1.up_2.weight. Drop parameter dla_up.ida_1.node_2.actf.0.weight. Drop parameter dla_up.ida_1.node_2.actf.0.bias. Drop parameter dla_up.ida_1.node_2.actf.0.running_mean. Drop parameter dla_up.ida_1.node_2.actf.0.running_var. Drop parameter dla_up.ida_1.node_2.actf.0.num_batches_tracked. Drop parameter dla_up.ida_1.node_2.conv.weight. Drop parameter dla_up.ida_1.node_2.conv.bias. Drop parameter dla_up.ida_1.node_2.conv.conv_offset_mask.weight. Drop parameter dla_up.ida_1.node_2.conv.conv_offset_mask.bias. Drop parameter dla_up.ida_2.proj_1.actf.0.weight. Drop parameter dla_up.ida_2.proj_1.actf.0.bias. Drop parameter dla_up.ida_2.proj_1.actf.0.running_mean. Drop parameter dla_up.ida_2.proj_1.actf.0.running_var. Drop parameter dla_up.ida_2.proj_1.actf.0.num_batches_tracked. Drop parameter dla_up.ida_2.proj_1.conv.weight. Drop parameter dla_up.ida_2.proj_1.conv.bias. Drop parameter dla_up.ida_2.proj_1.conv.conv_offset_mask.weight. Drop parameter dla_up.ida_2.proj_1.conv.conv_offset_mask.bias. Drop parameter dla_up.ida_2.up_1.weight. Drop parameter dla_up.ida_2.node_1.actf.0.weight. Drop parameter dla_up.ida_2.node_1.actf.0.bias. Drop parameter dla_up.ida_2.node_1.actf.0.running_mean. Drop parameter dla_up.ida_2.node_1.actf.0.running_var. Drop parameter dla_up.ida_2.node_1.actf.0.num_batches_tracked. Drop parameter dla_up.ida_2.node_1.conv.weight. Drop parameter dla_up.ida_2.node_1.conv.bias. Drop parameter dla_up.ida_2.node_1.conv.conv_offset_mask.weight. Drop parameter dla_up.ida_2.node_1.conv.conv_offset_mask.bias. Drop parameter dla_up.ida_2.proj_2.actf.0.weight. Drop parameter dla_up.ida_2.proj_2.actf.0.bias. Drop parameter dla_up.ida_2.proj_2.actf.0.running_mean. Drop parameter dla_up.ida_2.proj_2.actf.0.running_var. Drop parameter dla_up.ida_2.proj_2.actf.0.num_batches_tracked. Drop parameter dla_up.ida_2.proj_2.conv.weight. Drop parameter dla_up.ida_2.proj_2.conv.bias. Drop parameter dla_up.ida_2.proj_2.conv.conv_offset_mask.weight. Drop parameter dla_up.ida_2.proj_2.conv.conv_offset_mask.bias. Drop parameter dla_up.ida_2.up_2.weight. Drop parameter dla_up.ida_2.node_2.actf.0.weight. Drop parameter dla_up.ida_2.node_2.actf.0.bias. Drop parameter dla_up.ida_2.node_2.actf.0.running_mean. Drop parameter dla_up.ida_2.node_2.actf.0.running_var. Drop parameter dla_up.ida_2.node_2.actf.0.num_batches_tracked. Drop parameter dla_up.ida_2.node_2.conv.weight. Drop parameter dla_up.ida_2.node_2.conv.bias. Drop parameter dla_up.ida_2.node_2.conv.conv_offset_mask.weight. Drop parameter dla_up.ida_2.node_2.conv.conv_offset_mask.bias. Drop parameter dla_up.ida_2.proj_3.actf.0.weight. Drop parameter dla_up.ida_2.proj_3.actf.0.bias. Drop parameter dla_up.ida_2.proj_3.actf.0.running_mean. Drop parameter dla_up.ida_2.proj_3.actf.0.running_var. Drop parameter dla_up.ida_2.proj_3.actf.0.num_batches_tracked. Drop parameter dla_up.ida_2.proj_3.conv.weight. Drop parameter dla_up.ida_2.proj_3.conv.bias. Drop parameter dla_up.ida_2.proj_3.conv.conv_offset_mask.weight. Drop parameter dla_up.ida_2.proj_3.conv.conv_offset_mask.bias. Drop parameter dla_up.ida_2.up_3.weight. Drop parameter dla_up.ida_2.node_3.actf.0.weight. Drop parameter dla_up.ida_2.node_3.actf.0.bias. Drop parameter dla_up.ida_2.node_3.actf.0.running_mean. Drop parameter dla_up.ida_2.node_3.actf.0.running_var. Drop parameter dla_up.ida_2.node_3.actf.0.num_batches_tracked. Drop parameter dla_up.ida_2.node_3.conv.weight. Drop parameter dla_up.ida_2.node_3.conv.bias. Drop parameter dla_up.ida_2.node_3.conv.conv_offset_mask.weight. Drop parameter dla_up.ida_2.node_3.conv.conv_offset_mask.bias. Drop parameter ida_up.proj_1.actf.0.weight. Drop parameter ida_up.proj_1.actf.0.bias. Drop parameter ida_up.proj_1.actf.0.running_mean. Drop parameter ida_up.proj_1.actf.0.running_var. Drop parameter ida_up.proj_1.actf.0.num_batches_tracked. Drop parameter ida_up.proj_1.conv.weight. Drop parameter ida_up.proj_1.conv.bias. Drop parameter ida_up.proj_1.conv.conv_offset_mask.weight. Drop parameter ida_up.proj_1.conv.conv_offset_mask.bias. Drop parameter ida_up.up_1.weight. Drop parameter ida_up.node_1.actf.0.weight. Drop parameter ida_up.node_1.actf.0.bias. Drop parameter ida_up.node_1.actf.0.running_mean. Drop parameter ida_up.node_1.actf.0.running_var. Drop parameter ida_up.node_1.actf.0.num_batches_tracked. Drop parameter ida_up.node_1.conv.weight. Drop parameter ida_up.node_1.conv.bias. Drop parameter ida_up.node_1.conv.conv_offset_mask.weight. Drop parameter ida_up.node_1.conv.conv_offset_mask.bias. Drop parameter ida_up.proj_2.actf.0.weight. Drop parameter ida_up.proj_2.actf.0.bias. Drop parameter ida_up.proj_2.actf.0.running_mean. Drop parameter ida_up.proj_2.actf.0.running_var. Drop parameter ida_up.proj_2.actf.0.num_batches_tracked. Drop parameter ida_up.proj_2.conv.weight. Drop parameter ida_up.proj_2.conv.bias. Drop parameter ida_up.proj_2.conv.conv_offset_mask.weight. Drop parameter ida_up.proj_2.conv.conv_offset_mask.bias. Drop parameter ida_up.up_2.weight. Drop parameter ida_up.node_2.actf.0.weight. Drop parameter ida_up.node_2.actf.0.bias. Drop parameter ida_up.node_2.actf.0.running_mean. Drop parameter ida_up.node_2.actf.0.running_var. Drop parameter ida_up.node_2.actf.0.num_batches_tracked. Drop parameter ida_up.node_2.conv.weight. Drop parameter ida_up.node_2.conv.bias. Drop parameter ida_up.node_2.conv.conv_offset_mask.weight. Drop parameter ida_up.node_2.conv.conv_offset_mask.bias. Skip loading parameter hm.0.weight, required shapetorch.Size([64, 256, 3, 3]), loaded shapetorch.Size([256, 64, 3, 3]). Skip loading parameter hm.0.bias, required shapetorch.Size([64]), loaded shapetorch.Size([256]). Skip loading parameter hm.2.weight, required shapetorch.Size([64, 64, 3, 3]), loaded shapetorch.Size([2, 256, 1, 1]). Skip loading parameter hm.2.bias, required shapetorch.Size([64]), loaded shapetorch.Size([2]). Skip loading parameter st.0.weight, required shapetorch.Size([64, 256, 3, 3]), loaded shapetorch.Size([256, 64, 3, 3]). Skip loading parameter st.0.bias, required shapetorch.Size([64]), loaded shapetorch.Size([256]). Skip loading parameter st.2.weight, required shapetorch.Size([64, 64, 3, 3]), loaded shapetorch.Size([8, 256, 1, 1]). Skip loading parameter st.2.bias, required shapetorch.Size([64]), loaded shapetorch.Size([8]). Skip loading parameter wh.0.weight, required shapetorch.Size([64, 256, 3, 3]), loaded shapetorch.Size([256, 64, 3, 3]). Skip loading parameter wh.0.bias, required shapetorch.Size([64]), loaded shapetorch.Size([256]). Skip loading parameter wh.2.weight, required shapetorch.Size([64, 64, 3, 3]), loaded shapetorch.Size([8, 256, 1, 1]). Skip loading parameter wh.2.bias, required shapetorch.Size([64]), loaded shapetorch.Size([8]). Skip loading parameter ax.0.weight, required shapetorch.Size([64, 256, 3, 3]), loaded shapetorch.Size([256, 64, 3, 3]). Skip loading parameter ax.0.bias, required shapetorch.Size([64]), loaded shapetorch.Size([256]). Skip loading parameter ax.2.weight, required shapetorch.Size([64, 64, 3, 3]), loaded shapetorch.Size([256, 256, 1, 1]). Skip loading parameter ax.2.bias, required shapetorch.Size([64]), loaded shapetorch.Size([256]). Skip loading parameter cr.0.weight, required shapetorch.Size([64, 256, 3, 3]), loaded shapetorch.Size([256, 64, 3, 3]). Skip loading parameter cr.0.bias, required shapetorch.Size([64]), loaded shapetorch.Size([256]). Skip loading parameter cr.2.weight, required shapetorch.Size([64, 64, 3, 3]), loaded shapetorch.Size([256, 256, 1, 1]). Skip loading parameter cr.2.bias, required shapetorch.Size([64]), loaded shapetorch.Size([256]). Skip loading parameter reg.0.weight, required shapetorch.Size([64, 256, 3, 3]), loaded shapetorch.Size([256, 64, 3, 3]). Skip loading parameter reg.0.bias, required shapetorch.Size([64]), loaded shapetorch.Size([256]). Skip loading parameter reg.2.weight, required shapetorch.Size([2, 64, 1, 1]), loaded shapetorch.Size([2, 256, 1, 1]). No param conv1.weight. No param bn1.weight. No param bn1.bias. No param bn1.running_mean. No param bn1.running_var. No param bn1.num_batches_tracked. No param layer1.0.conv1.weight. No param layer1.0.conv1.bias. No param layer1.0.bn1.weight. No param layer1.0.bn1.bias. No param layer1.0.bn1.running_mean. No param layer1.0.bn1.running_var. No param layer1.0.bn1.num_batches_tracked. No param layer1.0.conv2.weight. No param layer1.0.conv2.bias. No param layer1.0.bn2.weight. No param layer1.0.bn2.bias. No param layer1.0.bn2.running_mean. No param layer1.0.bn2.running_var. No param layer1.0.bn2.num_batches_tracked. No param layer1.0.downsample.0.weight. No param layer1.0.downsample.1.weight. No param layer1.0.downsample.1.bias. No param layer1.0.downsample.1.running_mean. No param layer1.0.downsample.1.running_var. No param layer1.0.downsample.1.num_batches_tracked. No param layer1.1.conv1.weight. No param layer1.1.conv1.bias. No param layer1.1.bn1.weight. No param layer1.1.bn1.bias. No param layer1.1.bn1.running_mean. No param layer1.1.bn1.running_var. No param layer1.1.bn1.num_batches_tracked. No param layer1.1.conv2.weight. No param layer1.1.conv2.bias. No param layer1.1.bn2.weight. No param layer1.1.bn2.bias. No param layer1.1.bn2.running_mean. No param layer1.1.bn2.running_var. No param layer1.1.bn2.num_batches_tracked. No param layer2.0.conv1.weight. No param layer2.0.conv1.bias. No param layer2.0.bn1.weight. No param layer2.0.bn1.bias. No param layer2.0.bn1.running_mean. No param layer2.0.bn1.running_var. No param layer2.0.bn1.num_batches_tracked. No param layer2.0.conv2.weight. No param layer2.0.conv2.bias. No param layer2.0.bn2.weight. No param layer2.0.bn2.bias. No param layer2.0.bn2.running_mean. No param layer2.0.bn2.running_var. No param layer2.0.bn2.num_batches_tracked. No param layer2.0.downsample.0.weight. No param layer2.0.downsample.1.weight. No param layer2.0.downsample.1.bias. No param layer2.0.downsample.1.running_mean. No param layer2.0.downsample.1.running_var. No param layer2.0.downsample.1.num_batches_tracked. No param layer2.1.conv1.weight. No param layer2.1.conv1.bias. No param layer2.1.bn1.weight. No param layer2.1.bn1.bias. No param layer2.1.bn1.running_mean. No param layer2.1.bn1.running_var. No param layer2.1.bn1.num_batches_tracked. No param layer2.1.conv2.weight. No param layer2.1.conv2.bias. No param layer2.1.bn2.weight. No param layer2.1.bn2.bias. No param layer2.1.bn2.running_mean. No param layer2.1.bn2.running_var. No param layer2.1.bn2.num_batches_tracked. No param layer3.0.conv1.weight. No param layer3.0.conv1.bias. No param layer3.0.bn1.weight. No param layer3.0.bn1.bias. No param layer3.0.bn1.running_mean. No param layer3.0.bn1.running_var. No param layer3.0.bn1.num_batches_tracked. No param layer3.0.conv2.weight. No param layer3.0.conv2.bias. No param layer3.0.bn2.weight. No param layer3.0.bn2.bias. No param layer3.0.bn2.running_mean. No param layer3.0.bn2.running_var. No param layer3.0.bn2.num_batches_tracked. No param layer3.0.downsample.0.weight. No param layer3.0.downsample.1.weight. No param layer3.0.downsample.1.bias. No param layer3.0.downsample.1.running_mean. No param layer3.0.downsample.1.running_var. No param layer3.0.downsample.1.num_batches_tracked. No param layer3.1.conv1.weight. No param layer3.1.conv1.bias. No param layer3.1.bn1.weight. No param layer3.1.bn1.bias. No param layer3.1.bn1.running_mean. No param layer3.1.bn1.running_var. No param layer3.1.bn1.num_batches_tracked. No param layer3.1.conv2.weight. No param layer3.1.conv2.bias. No param layer3.1.bn2.weight. No param layer3.1.bn2.bias. No param layer3.1.bn2.running_mean. No param layer3.1.bn2.running_var. No param layer3.1.bn2.num_batches_tracked. No param layer4.0.conv1.weight. No param layer4.0.conv1.bias. No param layer4.0.bn1.weight. No param layer4.0.bn1.bias. No param layer4.0.bn1.running_mean. No param layer4.0.bn1.running_var. No param layer4.0.bn1.num_batches_tracked. No param layer4.0.conv2.weight. No param layer4.0.conv2.bias. No param layer4.0.bn2.weight. No param layer4.0.bn2.bias. No param layer4.0.bn2.running_mean. No param layer4.0.bn2.running_var. No param layer4.0.bn2.num_batches_tracked. No param layer4.0.downsample.0.weight. No param layer4.0.downsample.1.weight. No param layer4.0.downsample.1.bias. No param layer4.0.downsample.1.running_mean. No param layer4.0.downsample.1.running_var. No param layer4.0.downsample.1.num_batches_tracked. No param layer4.1.conv1.weight. No param layer4.1.conv1.bias. No param layer4.1.bn1.weight. No param layer4.1.bn1.bias. No param layer4.1.bn1.running_mean. No param layer4.1.bn1.running_var. No param layer4.1.bn1.num_batches_tracked. No param layer4.1.conv2.weight. No param layer4.1.conv2.bias. No param layer4.1.bn2.weight. No param layer4.1.bn2.bias. No param layer4.1.bn2.running_mean. No param layer4.1.bn2.running_var. No param layer4.1.bn2.num_batches_tracked. No param adaption3.weight. No param adaption2.weight. No param adaption1.weight. No param adaption0.weight. No param adaptionU1.weight. No param deconv_layers1.0.weight. No param deconv_layers1.1.weight. No param deconv_layers1.1.bias. No param deconv_layers1.1.running_mean. No param deconv_layers1.1.running_var. No param deconv_layers1.1.num_batches_tracked. No param deconv_layers2.0.weight. No param deconv_layers2.1.weight. No param deconv_layers2.1.bias. No param deconv_layers2.1.running_mean. No param deconv_layers2.1.running_var. No param deconv_layers2.1.num_batches_tracked. No param deconv_layers3.0.weight. No param deconv_layers3.1.weight. No param deconv_layers3.1.bias. No param deconv_layers3.1.running_mean. No param deconv_layers3.1.running_var. No param deconv_layers3.1.num_batches_tracked. No param deconv_layers4.0.weight. No param deconv_layers4.1.weight. No param deconv_layers4.1.bias. No param deconv_layers4.1.running_mean. No param deconv_layers4.1.running_var. No param deconv_layers4.1.num_batches_tracked. No param ax.4.weight. No param ax.4.bias. No param ax.6.weight. No param ax.6.bias. No param ax.8.weight. No param ax.8.bias. No param cr.4.weight. No param cr.4.bias. No param cr.6.weight. No param cr.6.bias. No param cr.8.weight. No param cr.8.bias. No param hm.4.weight. No param hm.4.bias. No param hm.6.weight. No param hm.6.bias. No param hm.8.weight. No param hm.8.bias. No param st.4.weight. No param st.4.bias. No param st.6.weight. No param st.6.bias. No param st.8.weight. No param st.8.bias. No param wh.4.weight. No param wh.4.bias. No param wh.6.weight. No param wh.6.bias. No param wh.8.weight. No param wh.8.bias. loaded ../dir_of_ckpt/ckpt_wireless/processor_best.pth, epoch 100 0%| | 0/2 [00:00<?, ?it/s](3000, 5) 50%|████████████████████████████████████████████▌ | 1/2 [00:00<00:00, 2.30it/s](3000, 5) 100%|█████████████████████████████████████████████████████████████████████████████████████████| 2/2 [00:00<00:00, 2.21it/s]
查看保存的文件，发现还是原始图片，表格结构完全没有识别。
另外，请问是哪里的问题？预训练模型从链接中下载。

Links to pretrained weights

Excited to test this TSR approach. Looks like the links to the pretrained weights are missing?

Paper & code of GeoLayoutLM?

Hi authors,
Thanks for the great repo!
Could you provide the paper link to Geolayoutlm?
Furthermore, I saw the code in the current release is the same as the code in the main branch. Can you update this?

Thank you so much!