adursun / wsddn.pytorch Goto Github PK

View Code? Open in Web Editor NEW

48.0 4.0 10.0 81 KB

Implementation of Weakly Supervised Deep Detection Networks using the latest version of PyTorch

License: Apache License 2.0

Dockerfile 1.77% Python 90.83% Shell 7.39%

deep-learning weakly-supervised-detection wsddn pytorch pascal-voc computer-vision

wsddn.pytorch's Introduction

WSDDN PyTorch

Implementation of Weakly Supervised Deep Detection Networks using the latest version of PyTorch.

Bilen, H., & Vedaldi, A. (2016). Weakly supervised deep detection networks. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (pp. 2846-2854).

Implementation Differences

Adam optimizer (instead of SGD)
Spatial regulariser isn't added

Experiments

VGG16 based model is closest to EB + Box Sc. case with model L, which reported 30.4 mAP in the paper
AlexNet based model is closest to EB + Box Sc. case with model S, which reported 33.4 mAP in the paper
Results when VGG16 is used as base model

aero	bike	bird	boat	bottle	bus	car	cat	chair	cow	table	dog	horse	mbike	person	plant	sheep	sofa	train	tv	mean
41.4	46.3	22.7	24.5	13.6	57.7	49.9	31.1	7.5	31.1	24.3	25.9	38.7	53.5	7.2	13.9	31.1	38.6	48.3	39.0	32.3

Results when AlexNet is used as base model

aero	bike	bird	boat	bottle	bus	car	cat	chair	cow	table	dog	horse	mbike	person	plant	sheep	sofa	train	tv	mean
38.1	41.5	27.1	18.6	10.3	48.8	47.6	36.8	1.6	25.9	28.5	30.4	39.7	46.8	15.1	12.4	28.3	32.4	44.2	44.8	30.9

Requirements

Docker (19.03.2)
nvidia-container-toolkit (https://github.com/NVIDIA/nvidia-docker)

Build Steps

git clone [email protected]:adursun/wsddn.pytorch.git
cd wsddn.pytorch
./prepare.sh
docker run --rm --gpus all --ipc=host -v `pwd`:/ws -it wsddn.pytorch /bin/bash

Training Steps

# for VGG based model
python src/train.py --base_net vgg

Evaluation Steps

# for VGG based model
# run `wget "https://www.dropbox.com/s/xyi4hgms6y3ldmj/vgg_epoch_20.pt?dl=1" -P states/` to use pretrained weights
python src/evaluate.py --base_net vgg --state_path states/vgg_epoch_20.pt

wsddn.pytorch's People

Contributors

Stargazers

Watchers

Forkers

bityangke hyzcn meteora9479 jayparanjape cynthiacoding tmlabonte ztonege putphapon qu-zhenyu yuzima

wsddn.pytorch's Issues

How to train the network without Ens?

Hello, how does the network proposed in this paper realize Ensemble Learning（Ens）? How can I train the network without Ensemble Learning（Ens）?

Use argparse instead of hardcoded parameters

Update Dockerfile to contain big requirements

running result?

hello. could you tell me your runing result (MAP) ?

I encountered a problem when I adjusted the batchsize to 2

Thank you for sharing your code,but wen i tried to adjust the batchsize to 2,i got this:
Traceback (most recent call last):
File "src/train.py", line 84, in
) in tqdm(train_dl, f"Epoch {epoch}"):
File "/home/lzr/miniconda3/envs/open-mmlab/lib/python3.7/site-packages/tqdm/std.py", line 1129, in iter
for obj in iterable:
File "/home/lzr/miniconda3/envs/open-mmlab/lib/python3.7/site-packages/torch/utils/data/dataloader.py", line 345, in next
data = self._next_data()
File "/home/lzr/miniconda3/envs/open-mmlab/lib/python3.7/site-packages/torch/utils/data/dataloader.py", line 856, in _next_data
return self._process_data(data)
File "/home/lzr/miniconda3/envs/open-mmlab/lib/python3.7/site-packages/torch/utils/data/dataloader.py", line 881, in _process_data
data.reraise()
File "/home/lzr/miniconda3/envs/open-mmlab/lib/python3.7/site-packages/torch/_utils.py", line 394, in reraise
raise self.exc_type(msg)
RuntimeError: Caught RuntimeError in DataLoader worker process 0.
Original Traceback (most recent call last):
File "/home/lzr/miniconda3/envs/open-mmlab/lib/python3.7/site-packages/torch/utils/data/_utils/worker.py", line 178, in _worker_loop
data = fetcher.fetch(index)
File "/home/lzr/miniconda3/envs/open-mmlab/lib/python3.7/site-packages/torch/utils/data/_utils/fetch.py", line 47, in fetch
return self.collate_fn(data)
File "/home/lzr/miniconda3/envs/open-mmlab/lib/python3.7/site-packages/torch/utils/data/_utils/collate.py", line 79, in default_collate
return [default_collate(samples) for samples in transposed]
File "/home/lzr/miniconda3/envs/open-mmlab/lib/python3.7/site-packages/torch/utils/data/_utils/collate.py", line 79, in
return [default_collate(samples) for samples in transposed]
File "/home/lzr/miniconda3/envs/open-mmlab/lib/python3.7/site-packages/torch/utils/data/_utils/collate.py", line 64, in default_collate
return default_collate([torch.as_tensor(b) for b in batch])
File "/home/lzr/miniconda3/envs/open-mmlab/lib/python3.7/site-packages/torch/utils/data/_utils/collate.py", line 55, in default_collate
return torch.stack(batch, 0, out=out)
RuntimeError: invalid argument 0: Sizes of tensors must match except in dimension 0. Got 516 and 900 in dimension 2 at /pytorch/aten/src/TH/generic/THTensor.cpp:612

have you meet this problem?

Try detectron2's voc evaluator

Spatial Regulariser

Hi, did you add the Spatial Regulariser in this code, I can't find it

Explain available arguments

State file names should be more informative

They might contain date of training and hyperparameters.

Unify test and train datasets

Ignore small boxes

Boxes with edges smaller than 20 pixels is ignored both in training and evaluation.

https://github.com/hbilen/WSDDN/blob/master/core/wsddn_test.m#L236

Use state dict instead of model

Use absolute paths

Relative paths should be replaced by absolute ones.

For example: https://github.com/adursun/wsddn.pytorch/blob/master/src/datasets.py#L44

how to use the method in my own dataset, and the dataset is voc format

Hello, if I want to use this algorithm in my own data, a remote sensing image dataset. How should I modify the codes.
In addition, what algorithm should I choose to generate proposals and how to format the proposal files to match this detector.

Thank you very much for your reply.

请问 edge box 的代码在哪里？EdgeBoxesVOC2007 的矩阵数据如何生成

Add Jupyter to Dockerfile

Add src directory

RuntimeError: Error(s) in loading state_dict for WSDDN:

Hey bro, when I tried to run the command:
python3.7 src/evaluate.py --state_path="./states/epoch_20_with_scores.pt"

I got this:
Traceback (most recent call last):
File "src/evaluate.py", line 22, in
net.load_state_dict(torch.load(args.state_path))
File "/usr/local/python3/lib/python3.7/site-packages/torch/serialization.py", line 387, in load
return _load(f, map_location, pickle_module, **pickle_load_args)
File "/usr/local/python3/lib/python3.7/site-packages/torch/serialization.py", line 574, in _load
result = unpickler.load()
File "/usr/local/python3/lib/python3.7/site-packages/torch/serialization.py", line 537, in persistent_load
deserialized_objects[root_key] = restore_location(obj, location)
File "/usr/local/python3/lib/python3.7/site-packages/torch/serialization.py", line 119, in default_restore_location
result = fn(storage, location)
File "/usr/local/python3/lib/python3.7/site-packages/torch/serialization.py", line 99, in _cuda_deserialize
return storage_type(obj.size())
File "/usr/local/python3/lib/python3.7/site-packages/torch/cuda/init.py", line 599, in _lazy_new
return super(_CudaBase, cls).new(cls, *args, **kwargs)
KeyboardInterrupt

Would you like to tell me how to use the saved model args to do evaluate?

Add VGG based model support

TypeError: call() got an unexpected keyword argument 'image'

Hi！I'm trying to run train.py to train the network but I'm having this problem:
TypeError: call() got an unexpected keyword argument 'image'
and It seems that the problem is in this part of utils.py:

how can I solve the problem，I'm not sure if it's the code or me, thank you!

low map issue...

After getting 10th epoch model. The program evaluated this model and returned a series of results. But it seems that the results are very low...