talreiss / accurate-interpretable-vad Goto Github PK

Hey, there is no access to the shanghaitech poses under: extracted_features/shanghaitech/train/pose.npy

>$: git lfs fetch

fetch: Fetching reference refs/heads/master
batch response: This repository is over its data quota. Account responsible for LFS bandwidth should purchase more data packs to restore access.                                                                                
error: failed to fetch some objects from 'https://github.com/talreiss/Accurate-Interpretable-VAD.git/info/lfs'

The pose.npy file has the following content:

version https://git-lfs.github.com/spec/v1
oid sha256:1e4a775b56fc599baf4e1b24ef2c99c3c89ec09bd8957eb1de76f55fe35ba934
size 194913310

Is there another way to obtain the pose data? e.g. google drive or similar?

Hi!

As I delved deeper into the code, it seems that the provided pose features have no apparent correlation with the bounding boxes. Specifically, in a single frame with multiple detections, such as persons and vehicles, it's challenging to discern which pose corresponds to which person within the frame.

Is there a straightforward method to directly link each human pose to the corresponding individual identified by the detector?
Could you provide the poses in the same format as the CLIP/velocities where the relations are present?

Thank you

I want train with my data, how add extract pose feature to pipeline

I encountered such a problem when preprocessing

Traceback (most recent call last):
  File "pre_processing/flows.py", line 85, in <module>
    extracting_flows(dataset_name=args.dataset_name, root=root, train=args.train)
  File "pre_processing/flows.py", line 68, in extracting_flows
    pred_flow = flownet2(ims).data
  File "/home/tjut_xuchuokun/anaconda3/envs/pytorch-CAD/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1130, in _call_impl
    return forward_call(*input, **kwargs)
  File "/home/tjut_xuchuokun/Accurate-Interpretable-VAD/pre_processing/flownet_networks/flownet2_models.py", line 113, in forward
    flownetc_flow2 = self.flownetc(x)[0]
  File "/home/tjut_xuchuokun/anaconda3/envs/pytorch-CAD/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1130, in _call_impl
    return forward_call(*input, **kwargs)
  File "/home/tjut_xuchuokun/Accurate-Interpretable-VAD/pre_processing/flownet_networks/FlowNetC.py", line 89, in forward
    out_corr = self.corr(out_conv3a, out_conv3b)  # False
  File "/home/tjut_xuchuokun/anaconda3/envs/pytorch-CAD/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1130, in _call_impl
    return forward_call(*input, **kwargs)
  File "/home/tjut_xuchuokun/Accurate-Interpretable-VAD/pre_processing/flownet_networks/correlation_package/correlation.py", line 58, in forward
    result = CorrelationFunction.apply(input1, input2, self.pad_size, self.kernel_size, self.max_displacement, self.stride1, self.stride2, self.corr_multiply)
  File "/home/tjut_xuchuokun/Accurate-Interpretable-VAD/pre_processing/flownet_networks/correlation_package/correlation.py", line 24, in forward
    correlation_cuda.forward(input1, input2, rbot1, rbot2, output,
AttributeError: module 'correlation_cuda' has no attribute 'forward'

I think I have followed the Readme to build the lib file

Dear authors,
Hello, I have a question, in the reproduction of 2.Optical Flow when there are some problems. According to your environment, on my linux 1080ti, execute python pre_processing / flows.py-dataset_name = ped2-train

The error in correlation _ forward _ cuda _ kernel : invalid device function, 
runtimeError : CUDA call failed problem,

I guess it is a FlowNet2 problem, so go to the FlowNet2 official website to use the environment it provides, but the environment version is too low to be compatible with your code. Is there any good solution ?
I would appreciate it if you could answer. Thank you !

Hi,your work is excellent! I have a question for you.When I executed the command "python pre_processing/bboxes.py --dataset_name=ped2",the following error occurred.

/opt/conda/lib/python3.7/site-packages/torch/functional.py:478: UserWarning: torch.meshgrid: in an upcoming release, it will be required to pass the indexing argument. (Triggered internally at ../aten/src/ATen/native/TensorShape.cpp:2895.)
return _VF.meshgrid(tensors, **kwargs) # type: ignore[attr-defined]
60%|█████████████████████████████████████████████████████████████████████████████████████████████▏ | 1200/2010 [01:39<01:06, 12.10it/s]
Traceback (most recent call last):
File "pre_processing/bboxes.py", line 163, in
extract(dataset_root=root, dataset_name=args.dataset_name, train=args.train)
File "pre_processing/bboxes.py", line 125, in extract
batch, _ = dataset.getitem(idx)
File "./video_dataset.py", line 240, in getitem
cur_img = np.transpose(cv2.imread(self.frame_addresses[i]), [2, 0, 1])
File "<array_function internals>", line 6, in transpose
File "/opt/conda/lib/python3.7/site-packages/numpy/core/fromnumeric.py", line 658, in transpose
return _wrapfunc(a, 'transpose', axes)
File "/opt/conda/lib/python3.7/site-packages/numpy/core/fromnumeric.py", line 55, in _wrapfunc
return _wrapit(obj, method, *args, **kwds)
File "/opt/conda/lib/python3.7/site-packages/numpy/core/fromnumeric.py", line 44, in _wrapit
result = getattr(asarray(obj), method)(*args, **kwds)
ValueError: axes don't match array

Hope to get your answer，thanks！

Thanks for sharing your wonderful work.

I have a minor question.
Your paper on arXiv says that you used Mask R-CNN for detection bounding boxes in videos.
But as far as I know, it is used for segmentation.
Wasn't it Faster R-CNN ?

Could not find the file "train_clip_lengths.npy" in score_calibration.py

Hi, your work is excellent. I have a question about the anomaly score calculation.

In the paper, the anomaly score is calculated through Eq(5), but in the code, the anomaly score for ShanghaiTech does not contain the deep features component.

Your work has greatly inspired me, so I hope you can help me solve the above issue, thanks!

I tried executing avenue.sh and found that url http://101.32.75.151:8181/dataset/avenue.tar.gz is taking too long time to respond. Could you please update this with latest url? or point me to right url for downloading manually?

Hi, your work is excellent. Can you provide 17 dimensional skeleton detection code? I want to train my dataset.

best

Hello,

Great work! I was wondering if you also evaluated you method on Street Scene Benchmark?

Paper: https://openaccess.thecvf.com/content_WACV_2020/papers/Ramachandra_Street_Scene_A_new_dataset_and_evaluation_protocol_for_video_WACV_2020_paper.pdf

Dataset: https://www.merl.com/demos/video-anomaly-detection

Hi, your work is excellent. I have a question about the pose detection: whether the input of pose detection is normalized.
I am looking forward to your reply.

Hi, your work is excellent. I have a question about the pose detection: whether the input of pose detection is normalized.
I am looking forward to your reply.