search

tensorflow / hub

A library for transfer learning by reusing parts of TensorFlow models.
https://tensorflow.org/hub
Apache License 2.0
3.48k stars 1.66k forks source link

GPU memory usage is close to the limit in colab #246

Closed Malathi15 closed 5 years ago

Malathi15 commented 5 years ago

When I tried to run Action Recognition on the UCF101 Dataset Notebook https://github.com/tensorflow/hub/blob/master/examples/colab/action_recognition_with_tf_hub.ipynb in colab I got the result (input video size is below 2MB)

But if I gave input video more than 2 MB I got ResourceExhausted Error. screenshot 141

Even I terminate the notebook and start the new one I got the same error.

arnoegw commented 5 years ago

Colabs are "free compute" for demonstration purposes, but they do have resource limits (independent of TF Hub). Raw pixel data from videos is bigger than the encoded data in the file. For bigger input data, please take the colab as a starting point for how to use the module on your own hardware.

Multiple colabs run under your Google account share resources. From menu "Runtime", select "Manage sessions" to find (and possibly terminate) your other ongoing colab sessions.

Malathi15 commented 5 years ago

Thanks for your reply @arnoegw . As per your suggestion when I terminate the other colab sessions colab takes up to 8MB videos. Is it possible to run a live video in colab? If it's possible then How can I run a live video in colab with out any memory error? I'm using 4 Megapixel IP CCTV camera.

16cs19 commented 5 years ago

how you give an other input video to that sample_video, are you using your own dataset ?