svip-lab / ppgnet Goto Github PK

你好,我运行了您给的test.sh文件,输出的结果看着是随机的,

Can this algorithm be used to detect specific linear targets on the way? Not for all straight lines

When I run train.py in Ubuntu, I keep having problems with not finding commands

Hey, thanks for your nice work.
I wanna try your work to see whether it works in my pics. First, I need to train it, I started by train.sh.
But when I use tensorboard to visualize the junctions and lines, it seems that line_pred output nothing like below. Is that normal?

Thanks!

Hello, in root directory i did "git lfs pull", and my files in path "ckpt/backbone/"
decoder_epoch_20.pth has 162.5Mb
encoder_epoch_20.pth has 95Mb

When I run ./train.sh I have:

Loading weights for net_encoder @ ckpt/backbone/encoder_epoch_20.pth
Loading weights for net_decoder @ ckpt/backbone/decoder_epoch_20.pth
Traceback (most recent call last):
File "main.py", line 521, in
fire.Fire(LSDTrainer)
File "/home/user/anaconda3/envs/ppgnet/lib/python3.6/site-packages/fire/core.py", line 127, in Fire
component_trace = _Fire(component, args, context, name)
File "/home/user/anaconda3/envs/ppgnet/lib/python3.6/site-packages/fire/core.py", line 366, in _Fire
component, remaining_args)
File "/home/user/anaconda3/envs/ppgnet/lib/python3.6/site-packages/fire/core.py", line 542, in _CallCallable
result = fn(*varargs, **kwargs)
File "main.py", line 121, in init
pin_memory=True
File "/home/user/anaconda3/envs/ppgnet/lib/python3.6/site-packages/torch/utils/data/dataloader.py", line 176, in init
sampler = RandomSampler(dataset)
File "/home/user/anaconda3/envs/ppgnet/lib/python3.6/site-packages/torch/utils/data/sampler.py", line 66, in init
"value, but got num_samples={}".format(self.num_samples))

ValueError: num_samples should be a positive integer value, but got num_samples=0

Hello! I have gtx1050mobile with CUDA on my laptop. My ./train.sh contains:

python main.py
--exp-name line_weighted_wo_focal_junc --backbone resnet50
--backbone-kwargs '{"encoder_weights": "ckpt/backbone/encoder_epoch_20.pth", "decoder_weights": "ckpt/backbone/decoder_epoch_20.pth"}'
--dim-embedding 256 --junction-pooling-threshold 0.2
--junc-pooling-size 64 --attention-sigma 1.5 --block-inference-size 2
--data-root ./data/indoorDist/ --junc-sigma 3
--batch-size 2 --gpus 0,1,2,3 --num-workers 0 --resume-epoch latest
--is-train-junc True --is-train-adj True
--vis-junc-th 0.1 --vis-line-th 0.1
- train --end-epoch 9 --solver SGD --lr 0.2 --weight-decay 5e-4 --lambda-heatmap 1. --lambda-adj 5.
- train --end-epoch 15 --solver SGD --lr 0.02 --weight-decay 5e-4 --lambda-heatmap 1. --lambda-adj 10.
- train --end-epoch 30 --solver SGD --lr 0.002 --weight-decay 5e-4 --lambda-heatmap 1. --lambda-adj 10.
- end

After run I get:

Loading weights for net_encoder @ ckpt/backbone/encoder_epoch_20.pth
Loading weights for net_decoder @ ckpt/backbone/decoder_epoch_20.pth
find 481 jucntions.
Traceback (most recent call last):
File "main.py", line 548, in
fire.Fire(LSDTrainer)
File "/home/user/.virtualenvs/PPGNet/lib/python3.6/site-packages/fire/core.py", line 127, in Fire
component_trace = _Fire(component, args, context, name)
File "/home/user/.virtualenvs/PPGNet/lib/python3.6/site-packages/fire/core.py", line 366, in _Fire
component, remaining_args)
File "/home/user/.virtualenvs/PPGNet/lib/python3.6/site-packages/fire/core.py", line 542, in _CallCallable
result = fn(*varargs, **kwargs)
File "main.py", line 377, in train
self._train_epoch()
File "main.py", line 238, in _train_epoch
junctions_gt
File "/home/user/.virtualenvs/PPGNet/lib/python3.6/site-packages/torch/nn/modules/module.py", line 493, in call
result = self.forward(*input, **kwargs)
File "/home/user/.virtualenvs/PPGNet/lib/python3.6/site-packages/torch/nn/parallel/data_parallel.py", line 150, in forward
return self.module(*inputs[0], **kwargs[0])
File "/home/user/.virtualenvs/PPGNet/lib/python3.6/site-packages/torch/nn/modules/module.py", line 493, in call
result = self.forward(*input, **kwargs)
File "/home/user/Загрузки/PPGNet/PPGNet/models/lsd.py", line 202, in forward
junc_hm, junc_coords = self.junc_infer(feat)
File "/home/user/.virtualenvs/PPGNet/lib/python3.6/site-packages/torch/nn/modules/module.py", line 493, in call
result = self.forward(*input, **kwargs)
File "/home/user/Загрузки/PPGNet/PPGNet/models/graph.py", line 42, in forward
coord = coord[ind[:self.max_juncs]]
File "/home/user/.virtualenvs/PPGNet/lib/python3.6/site-packages/torch/tensor.py", line 458, in array
return self.numpy()
TypeError: can't convert CUDA tensor to numpy. Use Tensor.cpu() to copy the tensor to host memory first.

我只用该模型前半部分做图片上关键点的检测，但是效果很差，也看不到优化的可能，不知道怎么弄了

Hallo,
I know this might get overlooked since the repository is relatively old, but maybe someone has an idea what might be the solution to my problem.
Normally I use the pytorch/pytorch:0.4.1-cuda9-cudnn7-devel Docker environment for my training of the PPGNet, which worked perfectly fine up until this week. Since then I get the following Error-message:

Traceback (most recent call last):
  File "main.py", line 523, in <module>
    fire.Fire(LSDTrainer)
  File "/opt/conda/lib/python3.6/site-packages/fire/core.py", line 141, in Fire
    component_trace = _Fire(component, args, parsed_flag_args, context, name)
  File "/opt/conda/lib/python3.6/site-packages/fire/core.py", line 471, in _Fire
    target=component.__name__)
  File "/opt/conda/lib/python3.6/site-packages/fire/core.py", line 681, in _CallAndUpdateTrace
    component = fn(*varargs, **kwargs)
  File "main.py", line 360, in train
    self._train_epoch()
  File "main.py", line 211, in _train_epoch
    for i, batch in enumerate(data_loader):
  File "/opt/conda/lib/python3.6/site-packages/torch/utils/data/dataloader.py", line 314, in __next__
    batch = self.collate_fn([self.dataset[i] for i in indices])
  File "/opt/conda/lib/python3.6/site-packages/torch/utils/data/dataloader.py", line 314, in <listcomp>
    batch = self.collate_fn([self.dataset[i] for i in indices])
  File "/home/simon/PPGNet/data/sist_line.py", line 24, in __getitem__
    lg = LineGraph().load(os.path.join(self.data_root, self.img[item][:-4] + ".lg"))
  File "/home/simon/PPGNet/data/line_graph.py", line 31, in load
    data = pickle.load(f)
ModuleNotFoundError: No module named 'sklearn.neighbors._kd_tree'

I use the recommended packages, as it can be seen in the following pip list:

Package          Version
---------------- -----------
asn1crypto       0.24.0
backcall         0.1.0
beautifulsoup4   4.6.0
certifi          2018.4.16
cffi             1.11.5
chardet          3.0.4
conda            4.5.8
conda-build      3.12.0
cryptography     2.2.2
Cython           0.29.32
decorator        4.3.0
filelock         3.0.4
fire             0.4.0
glob2            0.6
idna             2.6
ipython          6.4.0
ipython_genutils 0.2.0
jedi             0.12.1
Jinja2           2.10
llvmlite         0.36.0
lxml             4.9.1
MarkupSafe       1.0
mkl-fft          1.0.4
mkl-random       1.0.1
numba            0.53.1
numpy            1.19.5
olefile          0.45.1
opencv-python    4.6.0.66
parso            0.3.1
pexpect          4.6.0
pickleshare      0.7.4
Pillow           5.2.0
pip              21.3.1
pkginfo          1.4.2
prompt-toolkit   1.0.15
protobuf         3.19.6
psutil           5.4.6
ptyprocess       0.6.0
pycosat          0.6.3
pycparser        2.18
Pygments         2.2.0
pyOpenSSL        18.0.0
PySocks          1.6.8
PyYAML           3.13
requests         2.18.4
ruamel_yaml      0.15.37
scikit-learn     0.19.2
scipy            1.1.0
setuptools       39.2.0
Shapely          1.8.5.post1
simplegeneric    0.8.1
six              1.11.0
tensorboardX     2.5.1
termcolor        1.1.0
torch            0.4.1
torchvision      0.2.1
traitlets        4.3.2
triangle         20220202
urllib3          1.22
wcwidth          0.1.7
wheel            0.31.1

Currently I don't know a solution. I tried conda environments instead of docker, no success. Fresh installation on a different system, no success. Newer version of torch, torchvision, scikit-learn, no success.

But it seems like I am not the only one with the problem, since here (https://githubhelp.com/qingyonghu/sensaturban/issues/18) the same issue is also discussed without a solution.

I have installed the exact package versions as listed in the README.md file. By the way, NumPy version 0.15.0 does not exist. I suspect this is a typo and the README should read NumPy 1.15.0. I have also downloaded the data from Google Drive, unzipped via 7z, then untarred the resulting file.

I have modified the train.sh file to point to the downloaded data. When running the script on an AWS EC2 instance (p2.xlarge), I receive the following error:

(pytorch_p36) [ec2-user@ip-xxx-xx-xx-xxx PPGNet]$ ./train.sh
Loading weights for net_encoder @ ckpt/backbone/encoder_epoch_20.pth
Loading weights for net_decoder @ ckpt/backbone/decoder_epoch_20.pth
start training epoch: 0
ERROR: Unexpected segmentation fault encountered in worker.
ERROR: Unexpected segmentation fault encountered in worker.
Traceback (most recent call last):
  File "main.py", line 521, in <module>
ERROR: Unexpected segmentation fault encountered in worker.
ERROR: Unexpected segmentation fault encountered in worker.
ERROR: Unexpected segmentation fault encountered in worker.
    fire.Fire(LSDTrainer)
  File "/home/ec2-user/anaconda3/envs/pytorch_p36/lib/python3.6/site-packages/fire/core.py", line 127, in Fire
    component_trace = _Fire(component, args, context, name)
  File "/home/ec2-user/anaconda3/envs/pytorch_p36/lib/python3.6/site-packages/fire/core.py", line 366, in _Fire
    component, remaining_args)
  File "/home/ec2-user/anaconda3/envs/pytorch_p36/lib/python3.6/site-packages/fire/core.py", line 542, in _CallCallable
    result = fn(*varargs, **kwargs)
  File "main.py", line 358, in train
    self._train_epoch()
  File "main.py", line 209, in _train_epoch
    for i, batch in enumerate(data_loader):
  File "/home/ec2-user/anaconda3/envs/pytorch_p36/lib/python3.6/site-packages/torch/utils/data/dataloader.py", line 582, in __next__
    return self._process_next_batch(batch)
  File "/home/ec2-user/anaconda3/envs/pytorch_p36/lib/python3.6/site-packages/torch/utils/data/dataloader.py", line 608, in _process_next_batch
    raise batch.exc_type(batch.exc_msg)
TypeError: Traceback (most recent call last):
  File "/home/ec2-user/anaconda3/envs/pytorch_p36/lib/python3.6/site-packages/torch/utils/data/_utils/worker.py", line 99, in _worker_loop
    samples = collate_fn([dataset[i] for i in batch_indices])
  File "/home/ec2-user/anaconda3/envs/pytorch_p36/lib/python3.6/site-packages/torch/utils/data/_utils/worker.py", line 99, in <listcomp>
    samples = collate_fn([dataset[i] for i in batch_indices])
  File "/home/ec2-user/ray/PPGNet/data/sist_line.py", line 24, in __getitem__
    lg = LineGraph().load(os.path.join(self.data_root, self.img[item][:-4] + ".lg"))
  File "/home/ec2-user/ray/PPGNet/data/line_graph.py", line 29, in load
    data = pickle.load(f)
  File "sklearn/neighbors/binary_tree.pxi", line 1166, in sklearn.neighbors.kd_tree.BinaryTree.__setstate__
  File "stringsource", line 653, in View.MemoryView.memoryview_cwrapper
  File "stringsource", line 348, in View.MemoryView.memoryview.__cinit__
TypeError: a bytes-like object is required, not 'code'

Error in atexit._run_exitfuncs:
Traceback (most recent call last):
  File "/home/ec2-user/anaconda3/envs/pytorch_p36/lib/python3.6/multiprocessing/popen_fork.py", line 28, in poll
    pid, sts = os.waitpid(self.pid, flag)
  File "/home/ec2-user/anaconda3/envs/pytorch_p36/lib/python3.6/site-packages/torch/utils/data/_utils/signal_handling.py", line 63, in handler
    _error_if_any_worker_fails()
RuntimeError: DataLoader worker (pid 11499) is killed by signal: Segmentation fault.

Searching on forums, they suggest that you should set num_workers=0. When I tried doing that, I simply get a segmentation fault without premise:

(pytorch_p36) [ec2-user@ip-xxx-xx-xx-xxx PPGNet]$ ./train.sh
Loading weights for net_encoder @ ckpt/backbone/encoder_epoch_20.pth
Loading weights for net_decoder @ ckpt/backbone/decoder_epoch_20.pth
start training epoch: 0
./train.sh: line 13: 11528 Segmentation fault      python main.py --exp-name line_weighted_wo_focal_junc --backbone resnet50 --backbone-kwargs '{"encoder_weights": "ckpt/backbone/encoder_epoch_20.pth", "decoder_weights": "ckpt/backbone/decoder_epoch_20.pth"}' --dim-embedding 256 --junction-pooling-threshold 0.2 --junc-pooling-size 64 --attention-sigma 1.5 --block-inference-size 128 --data-root ./indoorDist --junc-sigma 3 --batch-size 16 --gpus 0,1,2,3 --num-workers 0 --resume-epoch latest --is-train-junc True --is-train-adj True --vis-junc-th 0.1 --vis-line-th 0.1 - train --end-epoch 9 --solver SGD --lr 0.2 --weight-decay 5e-4 --lambda-heatmap 1. --lambda-adj 5. - train --end-epoch 15 --solver SGD --lr 0.02 --weight-decay 5e-4 --lambda-heatmap 1. --lambda-adj 10. - train --end-epoch 30 --solver SGD --lr 0.002 --weight-decay 5e-4 --lambda-heatmap 1. --lambda-adj 10. - end

I also suspect this may be due to the batch size, even though a p2.xlarge has K80 GPUs so this should theoretically handle the training as per the README.md instructions (requiring a GPU with at least 24 GB of RAM). I've tried reducing the batch size to 1 and still get the same error.

The above errors when looking at the original training script as is suggests that there is something wrong with the line graph (.lg) files and I'm not familiar with how these are structured to proceed any further. Any help would be appreciated in sorting this out.

Step 3. specify the dataset path in the train.sh script.

For example, I extract the archive to the root of the project (/indoorDist).

In train.sh:

python main.py
--exp-name line_weighted_wo_focal_junc --backbone resnet50
--backbone-kwargs '{"encoder_weights": "ckpt/backbone/encoder_epoch_20.pth", "decoder_weights": "ckpt/backbone/decoder_epoch_20.pth"}'
--dim-embedding 256 --junction-pooling-threshold 0.2
--junc-pooling-size 64 --attention-sigma 1.5 --block-inference-size 16
--data-root /data/path --junc-sigma 3
--batch-size 16 --gpus 0,1,2,3 --num-workers 10 --resume-epoch latest
--is-train-junc True --is-train-adj True
--vis-junc-th 0.1 --vis-line-th 0.1
- train --end-epoch 9 --solver SGD --lr 0.2 --weight-decay 5e-4 --lambda-heatmap 1. --lambda-adj 5.
- train --end-epoch 15 --solver SGD --lr 0.02 --weight-decay 5e-4 --lambda-heatmap 1. --lambda-adj 10.
- train --end-epoch 30 --solver SGD --lr 0.002 --weight-decay 5e-4 --lambda-heatmap 1. --lambda-adj 10.
- end

Tell me please where I have to set path in this script?

Thanks for your work. I met a problem during training stage.
In graph.py line 118, there is a error "Calculated padded input size per channel: (7). Kernel size: (8). Kernel size can't be greater than actual input size", could you please help me with that?

Thank you for your work, I have some problems in implementing your code, I hope you can help me solve it in your busy schedule, thank you

Thanks for your work.

Unfortunately, I don't have sufficient resources to train model myself.

Do you plan to release trained checkpoint model anytime soon? It would be good to provide some instructions on running the trained model on some test images.

请问论文里您说的自己做的那个数据集可以发布吗？

Hey, I noticed that you mentioned a new dataset which include both indoor and outdoor images. It's a wonderful work, and could help other researcher begin their work. Will your team release the dataset mentioned in the paper?

find 0 jucntions. find 0 jucntions. epoch: [1][1382/2500], lr: 0.2, time_total: 1.87, time_data: 0.01, time_net: 1.70, time_vis: 0.16, loss: 0.4219, loss_heatmap: 0.1099, loss_adj_mtx: 0.0624 find 0 jucntions. find 0 jucntions. epoch: [1][1383/2500], lr: 0.2, time_total: 1.87, time_data: 0.01, time_net: 1.70, time_vis: 0.16, loss: 0.4001, loss_heatmap: 0.1532, loss_adj_mtx: 0.0494 find 0 jucntions. find 0 jucntions. epoch: [1][1384/2500], lr: 0.2, time_total: 1.87, time_data: 0.01, time_net: 1.70, time_vis: 0.16, loss: 0.4026, loss_heatmap: 0.1498, loss_adj_mtx: 0.0506 find 0 jucntions. find 0 jucntions. epoch: [1][1385/2500], lr: 0.2, time_total: 1.87, time_data: 0.01, time_net: 1.70, time_vis: 0.16, loss: 0.4419, loss_heatmap: 0.1236, loss_adj_mtx: 0.0637 find 0 jucntions. find 0 jucntions. epoch: [1][1386/2500], lr: 0.2, time_total: 1.87, time_data: 0.01, time_net: 1.70, time_vis: 0.16, loss: 0.5633, loss_heatmap: 0.1879, loss_adj_mtx: 0.0751 find 0 jucntions. find 0 jucntions. epoch: [1][1387/2500], lr: 0.2, time_total: 1.87, time_data: 0.01, time_net: 1.70, time_vis: 0.16, loss: 0.3670, loss_heatmap: 0.1545, loss_adj_mtx: 0.0425 find 0 jucntions. find 0 jucntions. epoch: [1][1388/2500], lr: 0.2, time_total: 1.87, time_data: 0.01, time_net: 1.70, time_vis: 0.16, loss: 0.4504, loss_heatmap: 0.1039, loss_adj_mtx: 0.0693 find 0 jucntions. find 0 jucntions. epoch: [1][1389/2500], lr: 0.2, time_total: 1.87, time_data: 0.01, time_net: 1.70, time_vis: 0.16, loss: 0.4568, loss_heatmap: 0.0956, loss_adj_mtx: 0.0722 find 0 jucntions. find 0 jucntions. epoch: [1][1390/2500], lr: 0.2, time_total: 1.87, time_data: 0.01, time_net: 1.70, time_vis: 0.16, loss: 0.3710, loss_heatmap: 0.1334, loss_adj_mtx: 0.0475

Is this training properly? or If not, what should I do?

能把训练好的模型放出来吗，像测试一下效果

测试结果的可视化结果很差，出现大面积的污染，主要污染表现为R、G、B三通道存在通道数值达到255，不确定是什么原因导致？

I try to reproduce the result with 4 1080TI ,and i reduce the batch size=4,
The detailed configuration
python main.py \ --exp-name line_weighted_wo_focal_junc --backbone resnet50 \ --backbone-kwargs '{"encoder_weights": "ckpt/backbone/encoder_epoch_20.pth", "decoder_weights": "ckpt/backbone/decoder_epoch_20.pth"}' \ --dim-embedding 256 --junction-pooling-threshold 0.2 \ --junc-pooling-size 64 --attention-sigma 1.5 --block-inference-size 4 \ --data-root "data/indoorDist" --junc-sigma 3 \ --batch-size 4 --gpus 0,1,2,3 --num-workers 10 --resume-epoch latest \ --is-train-junc True --is-train-adj True \ --vis-junc-th 0.1 --vis-line-th 0.1 \ - train --end-epoch 9 --solver SGD --lr 0.2 --weight-decay 5e-4 --lambda-heatmap 1. --lambda-adj 5. \ - train --end-epoch 15 --solver SGD --lr 0.02 --weight-decay 5e-4 --lambda-heatmap 1. --lambda-adj 10. \ - train --end-epoch 30 --solver SGD --lr 0.002 --weight-decay 5e-4 --lambda-heatmap 1. --lambda-adj 10. \ - end

Traceback (most recent call last):
File "main.py", line 522, in
fire.Fire(LSDTrainer)
File ".local/lib/python3.6/site-packages/fire/core.py", line 138, in Fire
component_trace = _Fire(component, args, parsed_flag_args, context, name)
File ".local/lib/python3.6/site-packages/fire/core.py", line 471, in _Fire
target=component.name)
File ".local/lib/python3.6/site-packages/fire/core.py", line 675, in _CallAndUpdateTrace
component = fn(*varargs, **kwargs)
File "main.py", line 359, in train
self._train_epoch()
File "main.py", line 227, in _train_epoch
img, heatmap_gt, adj_mtx_gt, self.lambda_heatmap, self.lambda_adj, junctions_gt
File "envs/r0.3.0/lib/python3.6/site-packages/torch/nn/modules/module.py", line 493, in call
result = self.forward(*input, **kwargs)
File "envs/r0.3.0/lib/python3.6/site-packages/torch/nn/parallel/data_parallel.py", line 152, in forward
outputs = self.parallel_apply(replicas, inputs, kwargs)
File "envs/r0.3.0/lib/python3.6/site-packages/torch/nn/parallel/data_parallel.py", line 162, in parallel_apply
return parallel_apply(replicas, inputs, kwargs, self.device_ids[:len(replicas)])
File "envs/r0.3.0/lib/python3.6/site-packages/torch/nn/parallel/parallel_apply.py", line 83, in parallel_apply
raise output
File "envs/r0.3.0/lib/python3.6/site-packages/torch/nn/parallel/parallel_apply.py", line 59, in _worker
output = module(*input, **kwargs)
File "envs/r0.3.0/lib/python3.6/site-packages/torch/nn/modules/module.py", line 493, in call
result = self.forward(*input, **kwargs)
File "LSD/PPGNet/models/lsd.py", line 291, in forward
adj_matrix_pred, loss_adj = block_adj_infer(feat_adj)
File "envs/r0.3.0/lib/python3.6/site-packages/torch/nn/modules/module.py", line 493, in call
result = self.forward(*input, **kwargs)
File "LSD/PPGNet/models/common.py", line 281, in forward
output = submodule(output)
File "envs/r0.3.0/lib/python3.6/site-packages/torch/nn/modules/module.py", line 493, in call
result = self.forward(*input, **kwargs)
File "LSD/PPGNet/models/lsd.py", line 70, in forward
line_feat = self.line_pool(feat, junc_st, junc_ed)
File "envs/r0.3.0/lib/python3.6/site-packages/torch/nn/modules/module.py", line 493, in call
result = self.forward(*input, **kwargs)
File "LSD/PPGNet/models/graph.py", line 86, in forward
output = F.grid_sample(feat[int(bs)].view(1, ch, h, w).expand(num_st, ch, h, w), sample_grid)
File "envs/r0.3.0/lib/python3.6/site-packages/torch/nn/functional.py", line 2717, in grid_sample
return torch.grid_sampler(input, grid, mode_enum, padding_mode_enum)
RuntimeError: CUDA out of memory. Tried to allocate 44.00 MiB (GPU 0; 10.91 GiB total capacity; 1.53 GiB already allocated; 15.38 MiB free; 94.03 MiB cached)

作者您好，我按照步骤来运行train.sh，但是直接就给我报错了， File "main.py", line 98
if resume_epoch and os.path.isfile(os.path.join("ckpt", exp_name, f"train_states_{resume_epoch}.pth")):
^
SyntaxError: invalid syntax
这应该是这句话附近哪里符号或者空格之类的有误？下载之后我没有改动过代码，我也没检查出哪里有问题，不知道怎么解决？

_pickle.UnpicklingError: invalid load key, 'v'.

您好，感谢您的工作。我在训练的时候出现了WARNING:root:NaN or Inf found in input tensor.这个问题，训练完之后效果非常差，因为显卡不够只按照您的建议做了如下修改：
batch-size 1 block-inference-size 32 请问效果很差是什么原因造成的呢，期待您的回答。

Hello, is there a demo to run on an image for a set of images? I noticed test.sh but it seems to run the training process. Thank you.