Video is too short error and Model loading error

[Sidd1609]

Hi, I am trying to evaluate using trained 3DResnet models would it be possible to have your version of the trained model as the link provided for the 3DResnet model and the optical flow model is not available.

Please also provide a more detailed explanation of how to train our own network for both inside and outside.

[yaorong0921]

Hi there, please try the link again.

For training with the videos inside and outside: To reduce computational overhead, I presaved the features from two video streams and utilized multiple layers to process them for obtaining the ultimate outcomes. However, I'm currently searching for the script as it was stored on a server that I may lose access to. I'll share it once I found it.

Best wishes,

[Sidd1609]

Thanks, that would be really helpful.

On Fri, Mar 24, 2023 at 9:38 PM yaorong0921 @.***> wrote:

Hi there, please try the link again.

For training with the videos inside and outside: To reduce computational overhead, I presaved the features from two video streams and utilized multiple layers to process them for obtaining the ultimate outcomes. However, I'm currently searching for the script as it was stored on a server that I may lose access to. I'll share it once I found it.

Best wishes,

— Reply to this email directly, view it on GitHub https://github.com/yaorong0921/Driver-Intention-Prediction/issues/5#issuecomment-1483056493, or unsubscribe https://github.com/notifications/unsubscribe-auth/ALTC7V72FV7BJR4MIUE3VO3W5XBHXANCNFSM6AAAAAAWF7LIJM . You are receiving this because you authored the thread.Message ID: @.***>

[Sidd1609]

Yes, the link is working now. I will test the inference using the trained weights and get back to you. Please could you also provide scripts to manually train for our own version of the dataset.

[Sidd1609]

Hi, the zip folder of the optical flow images seems corrupted, I have tried to open it on mac and Linux but keep getting errors. PLease could you check the link.

[Sidd1609]

Please also provide information on how to get sample inference results using any data.

[Sidd1609]

Do you have any inference scripts, where I could upload an inside and corresponding outside video to the model for predicting output and saving the results?

[ae208gpu]

@yaorong0921 Just following up on the inference script, is it possible to run the inference for getting results like the gif file in ReadMe?

[ae208gpu]

@yaorong0921 Getting the same issue for flownet images

Video is too short: ModelData/flownet2_road_camera/lturn/20141019_091623_289_439/ Video is too short:

[yaorong0921]

Hi @ae208gpu ,

1. For the gif file, I run the inference to get the prediction and then establish the gif.
2. Could you please check whether the dataloader works? You get this error maybe because there is no files found in the folder? Did you also get similar error here?: https://github.com/yaorong0921/Driver-Intention-Prediction/issues/2

[ae208gpu]

@yaorong0921

1. Please would you share the inference code for getting the prediction like the gif file?
2. There are image files in the folder for both inside and I use outside images from the link that was provided for optical flow images.
3. Please would you consider also stating how to benchmark your model to evaluate on i) only inside, ii) only outside, and iii) both on inside and outside images.

Your reply at the earliest would be really helpful.

[ae208gpu]

@yaorong0921 I checked the data loader it works fine, but the size of the images in the folder is the issue please could provide some information on how to fix the issue ?

[yaorong0921]

Hi @ae208gpu, what error did you receive regarding the image size? I have not encountered such an error.

[ae208gpu]

Error: RuntimeError: Caught RuntimeError in DataLoader worker process 0. Original Traceback (most recent call last): File "/anaconda3/envs/DIP/lib/python3.8/site-packages/torch/utils/data/_utils/worker.py", line 308, in _worker_loop data = fetcher.fetch(index) File "/anaconda3/envs/DIP/lib/python3.8/site-packages/torch/utils/data/_utils/fetch.py", line 51, in fetch data = [self.dataset[idx] for idx in possibly_batched_index] File "/anaconda3/envs/DIP/lib/python3.8/site-packages/torch/utils/data/_utils/fetch.py", line 51, in data = [self.dataset[idx] for idx in possibly_batched_index] File "Driver-Intention-Prediction/datasets/Brain4cars.py", line 274, in getitem clip = torch.stack(clip, 0) RuntimeError: stack expects a non-empty TensorList

Video is too short: flownet2_road_camera/rturn/20141220_161745_1_41/ Video is too short: flownet2_road_camera/end_action/20141019_141117_911_1061/ Video is too short: flownet2_road_camera/end_action/20141019_171211_1042_1192/ Video is too short: flownet2_road_camera/end_action/20141019_165104_768_918/

File does not exists: flownet2_road_camera/end_action/20141019_161013_1127_1277/

-> Please would you share the inference code for getting the prediction like the gif file? -> There are image files in the folder for both inside and I use outside images from the link that was provided for optical flow images. -> Please would you consider also stating how to benchmark your model to evaluate on i) only inside, ii) only outside, and iii) both on inside and outside images.

@yaorong0921

[ae208gpu]

What is the total size of flownet image folder (the unzipped version)? @yaorong0921

[ae208gpu]

@yaorong0921 how to resolve this issue with flownet images, is it necessary for the flownet images to be inside the road_camera folder of brain4cars.

[ae208gpu]

@yaorong0921 getting this error when running bash-Conv.sh

RuntimeError: stack expects a non-empty TensorList /ssd_scratch/cvit/srisiddarth16541/flownet2_road_camera/lturn/20141115_103835_933_1083/image-0298.png