annusha/unsup_temp_embed

Unsupervised learning of action classes with continuous temporal embedding (CVPR'19)

unsup_temp_embed's Issues

what is MoF?

Hi, can you explain what is mean over frames (MoF) and why do you think it's a good metric to evaluate the accuracy of your result with?

Granularity levels of the 50 Salads dataset

Hi, I have a question regarding the granularity levels of the 50 Salads dataset.
In the paper you mentioned you evaluated on two granularity levels: mid-level with 17 subaction classes and eval-level with 9 subaction classes.
However in the ground truth folder you provided 3 mapping files: mapping.txt with 19 classes, mappingeval.txt with 12 classes and mappinghigh.txt with 5 classes, all classes including action_start and action_end.
I wonder which mapping files are used for the results of eval-level (MoF 35.5%) and mid-level (30.2%) given in the paper? Are action_start and action_end also counted as a subaction class?
Regards

missing of fs_train.py and fs_test.py

Hi, I noticed that the train and test script (data_utils/FS_utils/fs_train.py and data_utils/FS_utils/fs_test.py) for 50Salads dataset are missing in the repository. Could you also upload them? Thanks for the efforts.

Global Mode

Hi Annusha,
Thanks for your work.

I was wondering if you have time to complete the training for the global model which we don't train for a specific class?
Because I could not get the global branch to work.

What is the function of "resume_str" in line 99 of corpus.py?

Hi Anna,
Thank you for making the code available. I just had a small concern. I am trying to run the code by setting model_name='nothing' (using just the raw features). However, I am getting the following error from line 99 of corpus.py.

AttributeError: 'Namespace' object has no attribute 'resume_str'

I then checked that "resume_str" is not part of the namespace. Please let me know what "resume_str" is.
Thank you.

how to read the evaluation log?

for the log: test.global!bg_dim30_ep60_nm_lr0.001_mlp_size0_bf_global(2020-07-21 10/55/46.284291)

How can we find the breaksfast number in Table 4? (which should be 26.4% F-1 score, and 41.8% MoF)

and in table 8? (which should be 18.3% MoF)

I only find the
2020-07-21 11:02:24,633 - 10 - global_corpus.py - video_level_clustering - MoF val: 0.3189252336448598
for table 7.

Hungarian matching when #clusters is not equal to #classes

Hi Anna,

thanks a lot for uploading the code of pipeline of unsupervised learning with unknown activity names. I just want to make sure if I correctly understood your Hungarian matching in the case that the number of clusters is not equal to the number of class labels.
In the paper you mentioned “the frames of the leftover clusters are set to background”. I assume that the “background” here corresponds to the label -1, which is only defined on the YTI dataset.
In this case, when you match the 50 clusters to the 48 ground truth classes on Breakfast, the remaining 2 clusters are simply ignored in the evaluation, as there is no background defined on Breakfast. On the YTI, you set K=9 and K’=5, which correspond to 45 clusters. In this case, there won’t be any leftover clusters but only 3 leftover ground classes. Therefore, no frames will be labeled as background during the evaluation. Only when the number of clusters on the YTI is larger than 48 will the frames of leftover clusters be assigned with label -1.
Please correct me there is any misunderstanding. Thanks for the efforts.

Regards