mohammed-elkomy / two-stream-action-recognition Goto Github PK

I followed the process in the Google Colab, but there is no output in the end.
I cannot play the demo.mp4 and no prediction shows.
How can I fix the bug?

Hi,

Thank you for providing a wonderful repository.

My system requirements are quite low and I want to use a pre-trained action recognition model and fine tune it for my data.

Can you please suggest me a simple approach using your model?

"Where do I get your pre-trained action recognition model?"
"Finetuning process: how should I be extracting features and use the model using your code"

I am working with spatial stream, my tf version is 1.12, after some iteration it stop, i tried all the things but do't work at all, i also tried with tf 1.11, but do't work

Hi, thank you for your great work.

I have a small custom dataset with 2 classes and I have extracted RGB + flow images already. I would like to use your pre-trained models and do the transfer learning on my own dataset. Could you please tell me the steps I should be following?

Also, I have local GPU servers, so I don't need colab for devlopment. How can I modify your coab script to run it as general jupyter notebook?

Thank you.

I'm using my Colab to run Action Recognition Walkthrough.ipynb. I've signed in my Google Account. But an error occurs when I run the first cell. The following is the detail:

MessageError Traceback (most recent call last)
in ()
12 gauth.credentials = GoogleCredentials.get_application_default()
13 gauth.SaveCredentialsFile('cred_this_is_me.txt') # we need this to start with cred*
---> 14 files.download('cred_this_is_me.txt') # download and keep it safe you will need it later to identify your account
15 drive = GoogleDrive(gauth)

2 frames
/usr/local/lib/python3.6/dist-packages/google/colab/files.py in download(filename)
176 'port': port,
177 'path': _os.path.abspath(filename),
--> 178 'name': _os.path.basename(filename),
179 })

/usr/local/lib/python3.6/dist-packages/google/colab/output/_js.py in eval_js(script, ignore_result)
37 if ignore_result:
38 return
---> 39 return _message.read_reply_from_input(request_id)
40
41

/usr/local/lib/python3.6/dist-packages/google/colab/_message.py in read_reply_from_input(message_id, timeout_sec)
104 reply.get('colab_msg_id') == message_id):
105 if 'error' in reply:
--> 106 raise MessageError(reply['error'])
107 return reply.get('data', None)
108

MessageError: TypeError: Failed to fetch

I ran the first cell and derived the above information. So how can I get 'cred_this_is_me.txt' using colab or not using colab?

Hi! It's great repo, thanks! Much work has been done)
But I can not reproduce the result. For example, I take a model trained in 5 epochs from your Google cloud and continue learning. If we compare the 10 epochs that you have trained with my 10 epochs, there are significant changes. Your best result is 0.72, and my result is not much different from 0.57, obtained after 5 epochs.
Maybe there is something that I miss? Thanks

Hello, do you have RGB images and optical flow images extracted from UCF101 and HMDB51?
I have downloaded the images processed by Feichtenhofer using google colab but it seems the zip file is broken and I cannot get images. So could you share your data with me through google drive? Thanks a lot!

hello could this be test with others datasets ?

Hi @mohammed-elkomy . First i would like to say thank you for your great work. I have learnt a lot from your repo.
I have a question i would like to ask you. Is there code to integrate Recurrent network fusion into two stream motion and spatial network instead of average fusion. I would like to integrate it then compare it to Average fusion.