About HDD dataloader image/label correspondance

Hello ,

I'm tring to train on HDD dataset. I'm creating my own dataloader. for each image/video folder there is a corresponding target .npy file. However, the size of the target numpy array does not correspond to the number of frames of the images or the video for the same session_id. could you please provide more information about the correspondence of the target labels to the images/videos. Is it possible to provide the HDD dataloader you have used in this experiment.

Thank you,

Hello ,

I'm tring to train on HDD dataset. I'm creating my own dataloader. for each image/video folder there is a corresponding target .npy file. However, the size of the target numpy array does not correspond to the number of frames of the images or the video for the same session_id. could you please provide more information about the correspondence of the target labels to the images/videos. Is it possible to provide the HDD dataloader you have used in this experiment.

Thank you,

Hi, thanks for your attention to OadTR, the preprocess file is Here.

Thank you for your reply, what i didn't understand the correspondence between image, video and target which contain .npy files. if we take example HDD/Image/201704151347/ has 1381 images HDD/video/201704151347/ has 116987 frames HDD/target/201704151347/ has 11700 size

i want to know how the target is related to images or video for dataloader to extract the features, Thank you,

Thank you for your reply, what i didn't understand the correspondence between image, video and target which contain .npy files. if we take example HDD/Image/201704151347/ has 1381 images HDD/video/201704151347/ has 116987 frames HDD/target/201704151347/ has 11700 size

i want to know how the target is related to images or video for dataloader to extract the features, Thank you,

Hi, the total length of the video is fixed, you need to use the original label provided by the author to interpolate to get the label of the image you extract. Note that the label provided by the author is obtained at 3FPS.

Hi @wangxiang1230,

In the paper you mention that out of 137 sections, 100 section are used for training and 37 are used for evaluation. Could you please provide the list of files that correspond to training set and which ones correspond to evaluaton set?

Thanks

Hi @wangxiang1230,

In the paper you mention that out of 137 sections, 100 section are used for training and 37 are used for evaluation. Could you please provide the list of files that correspond to training set and which ones correspond to evaluaton set?

Thanks

Hi, the anno file is here