mingbohong / sspnet Goto Github PK

Thank you for your great work！
I am not very familiar with the structure of mmdet, which file is the supervision loss of CAM?

Hello, I use the erase_with_uncertain_dataset dataset in Tiny Person and the test results in your paper are different, which part of the dataset are you using?

mmdet/models/module/heatmap.py line 94

threshold = -bg_score_sorted[bg_num - 1]

It seems in extra annotations, all objects including those with 'uncertain', 'ignore' and 'is_dense' are used for training. Would you mind sharing the annotation file which use only normal objects as asked in the TOD challenge?

Hello everyone! I have the mistake
File "tools/train.py", line 189, in
main()
File "tools/train.py", line 133, in main
logger = get_root_logger(log_file=log_file, log_level=cfg.log_level)
File "/opt/conda/envs/ssp/lib/python3.7/site-packages/mmcv/utils/config.py", line 507, in getattr
return getattr(self._cfg_dict, name)
File "/opt/conda/envs/ssp/lib/python3.7/site-packages/mmcv/utils/config.py", line 48, in getattr
raise ex
AttributeError: 'ConfigDict' object has no attribute 'log_level'
please help me!

I download the pretrained weights of "faster_rcnn_r50_sspnet" and get the following results:

 Average Precision  (AP) @[ IoU=0.50      | area=   all | maxDets=200 ] = 0.6227
 Average Precision  (AP) @[ IoU=0.50      | area=  tiny | maxDets=200 ] = 0.5835
 Average Precision  (AP) @[ IoU=0.50      | area= tiny1 | maxDets=200 ] = 0.4437
 Average Precision  (AP) @[ IoU=0.50      | area= tiny2 | maxDets=200 ] = 0.6191
 Average Precision  (AP) @[ IoU=0.50      | area= tiny3 | maxDets=200 ] = 0.6693
 Average Precision  (AP) @[ IoU=0.50      | area= small | maxDets=200 ] = 0.7189
 Average Precision  (AP) @[ IoU=0.50      | area=reasonable | maxDets=200 ] = 0.7330

I also try to change the maxDets to 1000 and the results is:

 Average Precision  (AP) @[ IoU=0.50      | area=   all | maxDets=1000 ] = 0.6251
 Average Precision  (AP) @[ IoU=0.50      | area=  tiny | maxDets=1000 ] = 0.5858
 Average Precision  (AP) @[ IoU=0.50      | area= tiny1 | maxDets=1000 ] = 0.4492
 Average Precision  (AP) @[ IoU=0.50      | area= tiny2 | maxDets=1000 ] = 0.6222
 Average Precision  (AP) @[ IoU=0.50      | area= tiny3 | maxDets=1000 ] = 0.6719
 Average Precision  (AP) @[ IoU=0.50      | area= small | maxDets=1000 ] = 0.7188
 Average Precision  (AP) @[ IoU=0.50      | area=reasonable | maxDets=1000 ] = 0.7331

My test command is:

tools/dist_test.sh configs/sspnet/faster_rcnn_r50_sspnet_1x_coco.py epoch_10.pth 2 --eval bbox 

So, how can I get the performance AP50 60.87 in README.md? Does the AP50 stand for AP50_tiny? Thanks!

try:
from panopticapi.utils import rgb2id
except ImportError:
rgb2id = None

It seems that the Panopticapi.utils module is not found, but it can be run. Will this affect the results?

@MingboHong Hi. I submit this issue to ask what is the configuration of RetinaNet_SPPNet reported in your paper. Would you mind sharing it?

Thank you for your work!I want to reproduce your experiment result.

Hi, I find in your implementation, even with Faster-RCNN-FPN, the result is higher than the previous baseline of about 10 mAP. Could you please share the configuration of how you implement the baseline Faster-RCNN-FPN?

@MingboHong Hello. What is the image_prefix used for training? There is no folder named 'images' that is used in configs/base/coco_detection.py. When I use 'tiny_set/train/', the iod_overlaps used in the sampler is void sometimes.

Thanks for your codes!
I met a problem when used SSPNet to train my own dataset, specifically, it will trigger a TypeError as follow:

And i check the code, in line 112 of ic_base_sampler.py, the following codes are the reason:

sampling_result = SamplingResult(pos_inds, neg_inds, bboxes, gt_bboxes, assign_result, gt_flags, iod_max_overlaps+conf)
in which the class SamplingResult can only be initiated with seven arguments (including self), hence the iod_max_overlaps+conf will cause the error. And when I remove the last parameter, it stills trigger another error as follow:

this makes me confused. Look forward to hearing from you, thanks a lot :)

The faster_rcnn_r50_sspnet_1x_coco can be trained correctly, but when I inference the trained model then encountered the error below.

inference_detector(model, imagepath)

python inference_img.py
/home/ubuntu/mmdetection-2.19.0/mmdet/models/builder.py:53: UserWarning: train_cfg and test_cfg is deprecated, please specify them in model
'please specify them in model', UserWarning)
load checkpoint from local path: /home/ubuntu/fovea/sspnet/epoch_1.pth
/home/ubuntu/mmdetection-2.19.0/mmdet/datasets/utils.py:69: UserWarning: "ImageToTensor" pipeline is replaced by "DefaultFormatBundle" for batch inference. It is recommended to manually replace it in the test data pipeline in your config file.
'data pipeline in your config file.', UserWarning)
/home/ubuntu/anaconda3/envs/openmmlab2/lib/python3.7/site-packages/torch/nn/functional.py:718: UserWarning: Named tensors and all their associated APIs are an experimental feature and subject to change. Please do not use them for anything important until they are released as stable. (Triggered internally at /pytorch/c10/core/TensorImpl.h:1156.)
return torch.max_pool2d(input, kernel_size, stride, padding, dilation, ceil_mode)
Traceback (most recent call last):
File "inference_img.py", line 26, in
result = inference_detector(model, imagepath)
File "/home/ubuntu/mmdetection-2.19.0/mmdet/apis/inference.py", line 147, in inference_detector
results = model(return_loss=False, rescale=True, **data)
File "/home/ubuntu/anaconda3/envs/openmmlab2/lib/python3.7/site-packages/torch/nn/modules/module.py", line 1051, in _call_impl
return forward_call(*input, **kwargs)
File "/home/ubuntu/anaconda3/envs/openmmlab2/lib/python3.7/site-packages/mmcv/runner/fp16_utils.py", line 109, in new_func
return old_func(*args, **kwargs)
File "/home/ubuntu/mmdetection-2.19.0/mmdet/models/detectors/base.py", line 174, in forward
return self.forward_test(img, img_metas, **kwargs)
File "/home/ubuntu/mmdetection-2.19.0/mmdet/models/detectors/base.py", line 147, in forward_test
return self.simple_test(imgs[0], img_metas[0], **kwargs)
File "/home/ubuntu/mmdetection-2.19.0/mmdet/models/detectors/faster_rcnn_ssp.py", line 186, in simple_test
x = self.extract_feat(img)
File "/home/ubuntu/mmdetection-2.19.0/mmdet/models/detectors/faster_rcnn_ssp.py", line 70, in extract_feat
x, att = self.neck(x)
ValueError: too many values to unpack (expected 2)

My inference command is :python demo/image_demo.py 2.jpg configs/sspnet/faster_rcnn_r50_sspnet_1x_coco_new.py checkpoints/epoch_10.pth --device cuda:3

Traceback (most recent call last):
File "demo/image_demo.py", line 50, in
main(args)
File "demo/image_demo.py", line 30, in main
result = inference_detector(model, args.img)
File "/home/SSPNet-master/mmdet/apis/inference.py", line 129, in inference_detector
data = test_pipeline(data)
File "/home/SSPNet-master/mmdet/datasets/pipelines/compose.py", line 41, in call
data = t(data)
File "/home/SSPNet-master/mmdet/datasets/pipelines/loading.py", line 61, in call
corner = results['img_info']['corner']
KeyError: 'corner

How can I fix it? or How to inference a image?

@MingboHong Hello. I find you and your group did a brilliant job in the first tiny object detection challenge. I am wondering why the performance you got in that competition seems higher a lot than SSPNet.

Thank you very much for sharing your code. I want to ask a few questions. Which part of the code does "WNS" reflect? Where is La's loss? thanks

Hi, Could you give me some help? PLZ!!
Tiny_person dataset seems not be straight used in your project.
Could you please afford custom dataset? PlZ
Your help is important for me!!!!

where is the code of dice loss?

i don not find a file named train,or relevant file

Could you please release the experiment setup and the processed dataset? Thank you very much!

I want to know how to crop images into patches,thank you

mmdet/models/module/heatmap.py line81
threshold = -bg_score_sorted[bg_num - 1]

I want to ask, how do I get the MR in the paper