Add post_process_depth_estimation to image processors - Githubissues

huggingface / transformers

🤗 Transformers: State-of-the-art Machine Learning for Pytorch, TensorFlow, and JAX.

https://huggingface.co/transformers

Apache License 2.0

132.76k stars 26.46k forks source link

Add post_process_depth_estimation to image processors #30917

Open NielsRogge opened 4 months ago

NielsRogge commented 4 months ago

Feature request

The image processors of depth estimation models could benefit from a post_process_depth_estimation method, similar to the post_process_object_detection, post_process_semantic_segmentation methods.

This includes:

[ ] DPT
[ ] Depth Anything (and v2)
[ ] ZoeDepth (soon: #30136) => this one requires some special postprocessing logic as shown in this notebook

Motivation

People now need to manually interpolate the predicted depth to the original size of the image as seen here.

The interpolation could be replaced by a simple image_processor.post_process_depth_estimation method which postprocesses the predicted depth and interpolates it to the original image size.

Your contribution

I could work on this, but would be great if someone else can take this up.

the-neural-networker commented 4 months ago

@NielsRogge, I can take this up!

ms42402 commented 2 months ago

Hi, if no one is working on this issue, I can take this!

amyeroberts commented 2 months ago

@ms42402 there's no active PRs, so feel free to tackle this if you'd like!

anirudh2781998 commented 2 months ago

hey anybody working on this? i would like to try and solve this issue

amyeroberts commented 2 months ago

@anirudh2781998 As above, if there isn't an active PR open, then anyone is free to work on any issue/feature request