Need more details to preprocess data on my own data

[jalaxy33]

Hello, @HYPJUDY .

I want to use the D-SSAD on my own dataset. I follow the instruction you gave in the Preprocess Data by Yourself, and get the frames and flow images extracted from my own dataset by the TSN.

Now I am stuck in the step 3. I have put the data/gen_thumos14_anno.py in the root directory of the SSN and made some directory mismatch modification. I don't know what to do next, mainly because I don't know how to create my own 'proposal_list.txt'. I found that the SSN code is all about thumos14 and activitynet, and I feel struggled to adjust the code to my own dataset.

I am confused now and asking for help. So could you please show me more details about the SSN data-preprocessing step?

[HYPJUDY]

Hi @jadada , you can refer to the process of generating of proposal list in SSN. For example, you may need to generate a normalized proposal list firstly, as mentioned in the readme. To generate the normalized proposal list, (1) you can refer to the example of thumos14 dataset and read this code to understand the meaning of each value and the structure of proposal list. (2) you can compare the example of thumos14 dataset and the thumos14 annotations of thumos14 to understand their forms. Then you may imitate the way and transfer from thumos14 to your own dataset. For more details, I suggest you to ask help from SSN repo. You may refer to their issues panel too. Good luck!

[xianguo-dev]

@HYPJUDY Hi! Thanks for your kind reply here.

But I am still confused about this" normalized_proposal_list" .Does SSN provide code to generate it? Because I found gen_proposal_list.py uses the authors' own " normalized_proposal_list" files. I could not find any code to generate proposal list and am struggling to imitate its content.

However, the data is too large and it's impossible to type my own proposal list one by one manually .So I'd like to know i that is there any code provided to generate the " normalized_proposal_list" or we really have to generate it by ourselves?

Looking forward to your generous help!

[HYPJUDY]

Hi, I do not have that code. I tried to search the answers in their repos but failed. It seems that some other people (e.g., https://github.com/yjxiong/action-detection/issues/100, https://github.com/open-mmlab/mmaction/issues/102) also have the same requirement to generate the normalized proposal lists but didn't work out. Maybe the best way is to write the code by yourself or email the original author. FYI, A Description of the Proposal Files might helps.

[xianguo-dev]

Thanks anyway,I've watched through all the issues in SSN and mmaction.But I still can't find any useful information except Description of the Proposal Files,which helps a bit.

Thanks again for you've made me realize throughly that I should find other ways.

[HYPJUDY]

You are welcome! Wish you find a solution!

[xianguo-dev]

@HYPJUDY Hi!I have worked out the specific meaning of most of the data in“ normalized_proposal_list ”but I still could not understand the origin of num_prop.I could not understand the data structure as the following part in“ normalized_proposal_list ” :

video_test_0000021 1 1 0 29 0 0.0000 0.0000 0.3955 0.4200 0 0.0000 0.0000 0.1335 0.1702 0 0.0000 0.0000 0.7125 0.7492 0 0.0000 0.0000 0.3109 0.3405 0 0.0000 0.0000 0.3833 0.4322 0 0.0000 0.0000 0.9327 0.9633 0 0.0000 0.0000 0.3048 0.3466 0 0.0000 0.0000 0.3772 0.4383 0 0.0000 0.0000 0.7003 0.7615 0 0.0000 0.0000 0.1213 0.1825 0 0.0000 0.0000 0.3649 0.4506 0 0.0000 0.0000 0.6942 0.7737 0 0.0000 0.0000 0.2926 0.3588 0 0.0000 0.0000 0.1030 0.2069 0 0.0000 0.0000 0.3109 0.4322 0 0.0000 0.0000 0.2926 0.4506 0 0.0000 0.0000 0.2742 0.4628 0 0.0000 0.0000 0.6942 0.9633 0 0.0000 0.0000 0.1030 0.4628 0 0.0000 0.0000 0.6758 1.0000 0 0.0000 0.0000 0.3833 0.7492 0 0.0000 0.0000 0.3109 0.7492 0 0.0000 0.0000 0.3649 0.9633 0 0.0000 0.0000 0.1335 0.7615 0 0.0000 0.0000 0.2926 0.9633 0 0.0000 0.0000 0.1335 0.3466 0 0.0000 0.0000 0.1213 0.3588 0 0.0000 0.0000 0.1213 0.9633 0 0.0000 0.0000 0.1335 0.4383

It seems that it does not have any regulation and the description you advised me to look at also does not introduce these numbers.

And I suppose that these numbers are random and are used in the parts of training and testing(randomize samples and locate frames) .

Would you please give me some advice?Sorry to interrupt you.

[HYPJUDY]

Hi @xianguo-dev , is this (A Description of the Proposal Files ) what you are looking for?

[xianguo-dev]

Thanks for your reply!Right.And I don't understand the following part:

11 86 0.9045 0.9625 0.1277 1.0000 86 0.8943 0.8943 0.0000 1.0000 86 0.8349 0.9595 0.1915 1.0000 86 0.3650 0.9121 0.6277 1.0000 86 0.3302 0.9037 0.6596 1.0000 86 0.2954 0.8936 0.6915 1.0000 86 0.2606 0.8810 0.7234 1.0000 86 0.2886 1.0000 0.1277 0.3856 86 0.2345 1.0000 0.1436 0.3537 86 0.1804 1.0000 0.1596 0.3218 86 0.1263 1.0000 0.1915 0.3059

The problem is that why this video has 11 proposals and how to generate this part?I have watched the corresponding video and find nothing special during these proposals.

[HYPJUDY]

I think they are the ground truth annotations, labeld by annotators.