escorciav / daps Goto Github PK

In the code the non-maximum suppression threshold for inference is set to 0.7, while in the paper, that value is 0.3 (for action detection).

So, I assume action detection requires less sparse proposals? Thank you!

First of all， Thank you very much for sharing the codes.
Could you please share the training codes? I just want to know something about how to write the training codes but do not care whether the codes pass functional test.
At least the codes of loss function in the paper?
Thanks anyways!

When I did the step 5 of do-you-want-to-try, I found that the file of the C3D representation of a couple of videos was an invalid hdf5 because python said IOError: Unable to open file (File signature not found). I also checked the file with h5debug command and got messages below.

$ h5debug pca_c3d_fc7_thumos14.hdf5 
HDF5-DIAG: Error detected in HDF5 (1.8.17) thread 140267664561920:
  #000: H5F.c line 604 in H5Fopen(): unable to open file
    major: File accessibilty
    minor: Unable to open file
  #001: H5Fint.c line 1087 in H5F_open(): unable to read superblock
    major: File accessibilty
    minor: Read failed
  #002: H5Fsuper.c line 277 in H5F_super_read(): file signature not found
    major: File accessibilty
    minor: Not an HDF5 file
cannot open file

All the hdf5 files in this folder were also invalid.
If you have free time, please reupload these sample files.

Hi @escorciav ,
First Thanks for your useful code, it's really helpful. When I run tools/generate_proposals.py, I got output like this in csv file:

245 171 0.42824159235 lsvc020163
53 11 0.406542558764 lsvc020163
4 -4 0.382954354819 lsvc020163
27 -28 0.379393531972 lsvc020163

I wonder what does the negative numbers mean and can I delete these records from the proposal lists as in-normal cases? Could you please give some hints on this? Thanks in advance.

Hi, I can't install conda, how to run your code without conda?? Thanks.

I read the codes and find that a fc layer is utilized following the lstm module. However, I still wonder how does the output of the fc layer generate temporal proposals? Could you please make it concretely?

Thank you to share your code and contribution.

My problem is when I use data/sample/c3d_after_pca.hdf5, It doesn't work with printing some error message.

Command :
python tools/generate_proposals.py -iv video_test_0000541 -ic3d data/samples/c3d_after_pca.hdf5 -imd data/models/T512K64_thumos14.npz

Error line and its message :
with h5py.File(self.filename, 'r') as fobj: -> self.filename = data/samples/c3d_after_pca.hdf5
print IOError: IOError(...found)',)

Could you advise to me why does it happen?
Maybe, It is highly basic question but I don't have any experience to deal with .hdf5 file format.

Hi~
In your paper, i find you design an action classifier to evaluate mAP for proposals, but i don't find this code, would you mind tell me where is the code and how to use it ?
Thank you very much! Best wishes !

hi, authors, @escorciav I want to reduce the video dimension from (num_frames, ori_dimension) to (num_frames, new_dimension), and I find that in the readme, you give an example, like this, I want to know what the means about 'U', could you explain it? thanks!!