Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

The video quality obtained by inference is relatively poor #28

Open
huoguangdiandian opened this issue Jan 20, 2025 · 5 comments
Open

Comments

@huoguangdiandian
Copy link

Thank you very much for your work. After reasoning, I found that the video quality was relatively poor. I don't know what caused it
I am using the configuration file is configs/magicdrive/inference/fullx224x400_stdit3_CogVAE_boxTDS_wCT_xCE_wSST py, The validation_index=[0] and num_frames=129 are modified ,the val pkl of inputing, change the number of 'infos' to 1000, and 'scene_token' only keeps the first element in the list .
The reason for reducing the amount of input data is that the machine memory is too small, I am checking whether the input data is incomplete

@flymin
Copy link
Owner

flymin commented Jan 20, 2025

Changing metadata does not help with memory cost. You don’t have to do that.

Please be reminded that you are running the lowest resolution supported. It is normal that the quality does not catch up that of high-resolution ones.

@huoguangdiandian
Copy link
Author

Changing metadata does not help with memory cost. You don’t have to do that.

Please be reminded that you are running the lowest resolution supported. It is normal that the quality does not catch up that of high-resolution ones.

Thank you very much for your reply, I will use the high resolution configuration file to see the effect.
If the metadata is not changed, memory and gpu memory can be very expensive, For example, I use fullx848x1600_stdit3_CogVAE_boxTDS_wCT_xCE_wSST.py configuration file, what can I do to reduce memory and gpu memory usage

@flymin
Copy link
Owner

flymin commented Jan 20, 2025

@huoguangdiandian
Copy link
Author

metadata refers to the pkl files.

Please see https://github.com/flymin/MagicDriveDiT/blob/main/doc/FAQ.md#q21-minimum-gpu-memory-requirements-for-inference

I just read the link you sent, and the reasons for the poor cpu usage and video quality are probably clear
But there are two parameters whose meaning is not sure, the first one is sp, this parameter is sp_size? The second is the bit rate, this parameter I did not find in the configuration file and inference code, can you tell me where to set it? thank you

@flymin
Copy link
Owner

flymin commented Jan 21, 2025

When we output the results as an MP4 video, the bitrate is limited to 4M on this line

As indicated by the doc.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

2 participants