ranjaykrishna / sst Goto Github PK

Hi，ranjaykrishna：
I think there's one problem in data.py. In line 218, the default setting of argument 'replace' in np.random.choice() is True, which will lead repeated sampling windows. So in my opinion, the correct form should be:
sample = np.random.choice(nWindows, self.max_W, replace=False)
Though I don't how much it will affect the performance.

Hi! Thank you for your nice work!
It seems that it needs these two files in data to run the training.

Hey, bro! THX your code! I am new for deep learning, and its my first time to use torch-dataset! Cool!
Is there something bug in models.py ,line41
rnn_output, _ = self.rnn(features)
rnn should forward( input, h0)?
And I did something to translate this repository to working with python3.Add a branch?

Hi ranjaykrishna!
I am reading your code and I notice that in your code(train.py) you mentioned two arguments(labels & vid-ids). But I cannot find them in official dataset. Are they produced by data file(activity_net.v1-3.min.json)?

Would you share the code to produce these two files?

Hi , thanks very much for your efforts.
I noticed that this is not the official repo.
Could you please tell me what's the difference between them, such as implementation details and performance？
Thank you in advance！

Hi ranjaykrishna,

I love your excellent project, your code is really elegant.

I'm trying to reuse your code, but I met some problems with the batch size(by default that is 20), the batch size is too large for a single TITAN X, I have found for a single TITAN X, 2 is the max batch size. So do you use nn.dataparallel() or something else?

Any help is appreciated

Best Regards

I have found in the train.py the iou function
start_i, end_i = interval[0], interval[1]

intersection = max(0, min(end, end_i) - max(start, start_i))
union = min(max(end, end_i) - min(start, start_i), end - start + end_i - start_i)
overlap = float(intersection) / (union + 1e-8)

when the start time is negative, whether the iou should be 0 anyway?

Traceback (most recent call last):
File "D:/paper_with_code/SST-master/train.py", line 283, in
train(epoch, w1)
File "D:/paper_with_code/SST-master/train.py", line 240, in train
for batch_idx, (features, masks, labels) in enumerate(train_loader):
File "D:\Anaconda3\lib\site-packages\torch\utils\data\dataloader.py", line 278, in iter
return _MultiProcessingDataLoaderIter(self)
File "D:\Anaconda3\lib\site-packages\torch\utils\data\dataloader.py", line 682, in init
w.start()
File "D:\Anaconda3\lib\multiprocessing\process.py", line 112, in start
self._popen = self._Popen(self)
File "D:\Anaconda3\lib\multiprocessing\context.py", line 223, in _Popen
return _default_context.get_context().Process._Popen(process_obj)
File "D:\Anaconda3\lib\multiprocessing\context.py", line 322, in _Popen
return Popen(process_obj)
File "D:\Anaconda3\lib\multiprocessing\popen_spawn_win32.py", line 89, in init
reduction.dump(process_obj, to_child)
File "D:\Anaconda3\lib\multiprocessing\reduction.py", line 60, in dump
ForkingPickler(file, protocol).dump(obj)
File "D:\Anaconda3\lib\site-packages\h5py_hl\base.py", line 308, in getnewargs
raise TypeError("h5py objects cannot be pickled")
TypeError: h5py objects cannot be pickled

ranjaykrishna Do you know what's the problem here, thx.