google-deepmind / google-deepmind/c3_neural_compression

Video coding results

Open
#4 0 comments 0 reactions 0 assignees View on GitHub
Dominant language
Python
Stars
101
Forks
10
PR merge metrics
No merged PRs in 30d

Description

Hi 👋

I've been doing some video encoding using C3 and I got some weird results here and there. For the context, I am using the UVG dataset with the `configs/uvg.py` file with the following changes:

```python
# ... configs/uvg.py
exp.dataset.num_frames = 9 # I want to code only 9 frames
exp.opt.num_noise_steps = 10_000 # Faster encoding
exp.opt.max_num_ste_steps = 1_000 # Faster encoding

exp.dataset.video_idx = 4 # I want to code a single sequence e.g. the 4th video
exp.dataset.skip_examples = 0 # I specify explicitly the video_idx, so I can start at 0
exp.dataset.num_examples = 48 # A 1920x1080 frame has 48 non-overlapping 180x240 patches
```

Here are the results I obtained sequence-wise (ignore the average graph 😉). Some sequences (Bosphorus, YachtRide) failed to converge. Have I done something wrong with the parameters within the config file? Or is it due to the small patch size and the reduced number of frames which increase the relative share of the NN parameters in the overall rate?

![image](https://github.com/user-attachments/assets/e8b365a2-e698-4ef2-8974-cb644ff186a7)

Thanks!

Contributor guide

Open the contributing guide

Assessment

This issue has not been assessed yet.

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.