mzolfaghari / ECO-efficient-video-understanding

Code and models of paper " ECO: Efficient Convolutional Network for Online Video Understanding", ECCV 2018
MIT License
437 stars 96 forks source link

about the video_data_param and train.txt #7

Closed Ai-is-light closed 6 years ago

Ai-is-light commented 6 years ago

@mzolfaghari Thanks for your excellent repo and paper. video_data_param { source: ".../train.txt" batch_size: 16 new_length: 1 new_width: 320 new_height: 240 num_segments: 16 modality: RGB shuffle: true name_pattern: "%05d.jpg" }

I want to use your model to train my custom datasets, so, would you mind telling us the format in your train.txt. And, what's more, as for the num_segments: 16, there are 16 {mean_value: [104] mean_value: [117] mean_value: [123] } the number of the num_segments is the same as the {mean_value}?

mzolfaghari commented 6 years ago

Hi @Ai-is-light ,

Format for the data list is like this: /path_to_data_folder/ number_of_frames action_label

First column path to the data folder, the second column shows the number of frames for each video and last column is the label of action.

Yes, for each segment we need to provide the mean value (e.g. for 16 segments we provide 16 mean values)