KaiyangZhou / pytorch-vsumm-reinforce

Unsupervised video summarization with deep reinforcement learning (AAAI'18)
MIT License
472 stars 150 forks source link

0.0% Fscore for videos and probably wrong Summary. #76

Open TheMaveric opened 3 years ago

TheMaveric commented 3 years ago

I am getting 0.0 Fscore and the summaries generated are either 3 seconds long and wrong or theyre 30 seconds and vague (probably wrong too). epoch_reward_0 epoch_reward_1 log_test.txt log_train.txt overall_reward score_video_2 score_video_5 score_video_8 score_video_12 score_video_15

Below are the contents of the log_train.txt file.

Args:Namespace(beta=0.01, dataset='datasets/eccv16_dataset_summe_google_pool5.h5', evaluate=False, gamma=0.1, gpu='0', hidden_dim=256, input_dim=1024, lr=1e-05, max_epoch=60, metric='summe', num_episode=5, num_layers=1, resume='', rnn_cell='lstm', save_dir='log/summe-split0', save_results=False, seed=1, split='datasets/summe_splits.json', split_id=0, stepsize=30, use_cpu=False, verbose=True, weight_decay=1e-05)

Currently using CPU Initialize dataset datasets/eccv16_dataset_summe_google_pool5.h5

total videos 25. # train videos 20. # test videos 5

Initialize model Model size: 2.62605M ==> Start training epoch 1/60 reward 0.8963902491331102
epoch 2/60 reward 0.8972550570964813
epoch 3/60 reward 0.8970675492286683
epoch 4/60 reward 0.8971286928653719
epoch 5/60 reward 0.8963818806409837
epoch 6/60 reward 0.896572777032852
epoch 7/60 reward 0.8964594054222106
epoch 8/60 reward 0.896507331132889
epoch 9/60 reward 0.8974075722694396
epoch 10/60 reward 0.8957048553228377
epoch 11/60 reward 0.8968981993198394
epoch 12/60 reward 0.8965510278940201
epoch 13/60 reward 0.8972574228048324
epoch 14/60 reward 0.896853615641594
epoch 15/60 reward 0.8963686144351959
epoch 16/60 reward 0.8969772887229919
epoch 17/60 reward 0.8977362161874771
epoch 18/60 reward 0.8969273221492766
epoch 19/60 reward 0.8961024188995361
epoch 20/60 reward 0.8966317284107209
epoch 21/60 reward 0.8962920480966569
epoch 22/60 reward 0.8959967434406279
epoch 23/60 reward 0.8971632570028305
epoch 24/60 reward 0.8966683238744737
epoch 25/60 reward 0.8961001712083817
epoch 26/60 reward 0.8959629529714583
epoch 27/60 reward 0.8962142568826674
epoch 28/60 reward 0.896654149889946
epoch 29/60 reward 0.8972274744510651
epoch 30/60 reward 0.8969877260923385
epoch 31/60 reward 0.897369709610939
epoch 32/60 reward 0.8968648070096972
epoch 33/60 reward 0.8968213593959808
epoch 34/60 reward 0.8971809494495391
epoch 35/60 reward 0.8975461572408676
epoch 36/60 reward 0.8970616376399994
epoch 37/60 reward 0.896257193684578
epoch 38/60 reward 0.8968519711494446
epoch 39/60 reward 0.8967313235998153
epoch 40/60 reward 0.8973873049020767
epoch 41/60 reward 0.8970531791448593
epoch 42/60 reward 0.8969561916589738
epoch 43/60 reward 0.8972805547714232
epoch 44/60 reward 0.8977496469020844
epoch 45/60 reward 0.8977025932073592
epoch 46/60 reward 0.8980738395452498
epoch 47/60 reward 0.8963353443145753
epoch 48/60 reward 0.8971281045675278
epoch 49/60 reward 0.8968862169981003
epoch 50/60 reward 0.8971532964706421
epoch 51/60 reward 0.896593438386917
epoch 52/60 reward 0.8973712176084518
epoch 53/60 reward 0.8967011392116548
epoch 54/60 reward 0.8982456147670745
epoch 55/60 reward 0.897282282114029
epoch 56/60 reward 0.8965970069169996
epoch 57/60 reward 0.8967110335826873
epoch 58/60 reward 0.8953982496261597
epoch 59/60 reward 0.8963999301195145
epoch 60/60 reward 0.8975783979892731
==> Test


No. Video F-score 1 video_12 0.0% 2 video_15 55.1% 3 video_2 22.3% 4 video_5 29.7% 5 video_8 49.4%


Average F-score 31.3% Finished. Total elapsed time (h:m:s): 0:24:16 Model saved to log/summe-split0\model_epoch60.pth.tar

Below are the contents of the log_test.txt file.

Args:Namespace(beta=0.01, dataset='datasets/eccv16_dataset_summe_google_pool5.h5', evaluate=True, gamma=0.1, gpu='0', hidden_dim=256, input_dim=1024, lr=1e-05, max_epoch=60, metric='summe', num_episode=5, num_layers=1, resume='log/summe-split0/model_epoch60.pth.tar', rnn_cell='lstm', save_dir='log/summe-split0', save_results=True, seed=1, split='datasets/summe_splits.json', split_id=0, stepsize=30, use_cpu=False, verbose=True, weight_decay=1e-05)

Currently using CPU Initialize dataset datasets/eccv16_dataset_summe_google_pool5.h5

total videos 25. # train videos 20. # test videos 5

Initialize model Model size: 2.62605M Loading checkpoint from 'log/summe-split0/model_epoch60.pth.tar' Evaluate only ==> Test


No. Video F-score 1 video_12 0.0% 2 video_15 55.1% 3 video_2 22.3% 4 video_5 29.7% 5 video_8 49.4%


Average F-score 31.3%

As you can see video_12 has a Fscore of 0.0% and also the FScores vary drastically. I followed the instructions but I cant seem to figure out what I did wrong. Any help is much appreciated.

JosephineRabbit commented 3 years ago

Hi, I meet this issue too.