yjxiong/tsn-pytorch

weried phenomenon about using main.py train ucf101 dataset

dandingol03 opened this issue · 1 comments

Hi, @yjxiong
i use the main.py script to train ucf101 , but it seems like the script is just loading the rawframe, and never begin to train the model. Here is my setting:

python main.py ucf101 RGB <ucf101_rgb_train_list> <ucf101_rgb_val_list>
--arch BNInception --num_segments 3
--gd 20 --lr 0.001 --lr_steps 30 60 --epochs 80
-b 120 -j 8 --dropout 0.8
--snapshot_pref ucf101_bninception_

Thanks in advance~

i figure out why, because i use the google colab, it needs a long time to load data. After that, it begin to train