training

Hello, I saw in the article that each video is divided into segments with a fixed length of 60, and when the FPS is 30, the sum of these segments is 2 seconds. How to set this up in code.

You can modify this setting at:

LRNet/training/configs/args_train.yaml

Line 1 in 408f288

BLOCK_SIZE: 30

If you change the segment length in train-setting, using the same length in the test-setting would have better results:

LRNet/training/configs/args_test.yaml

Line 1 in 408f288

BLOCK_SIZE: 30

The setting of 60 frames reported in the artical might be out-of-date. Our subsequent experiments found that 30 frames could reach slightly better performance.

If set to 60, do other hyperparameters need to be changed？

Also, it is used during test evaluation. For example, if a video has 303 frames, it will be cut into 10 samples (each sample contains 30 consecutive frames), and the last 3 frames will be discarded. Is there such a rule?

If set to 60, do other hyperparameters need to be changed？

None of the other hyperparameters need to be changed.

Also, it is used during test evaluation. For example, if a video has 303 frames, it will be cut into 10 samples (each sample contains 30 consecutive frames), and the last 3 frames will be discarded. Is there such a rule?

Yes, the codes

LRNet/training/evaluate.py

Line 61 in 408f288

dataset.load_data_test_all(BLOCK_SIZE, BATCH_SIZE)

would handle it.