search

lorjul / fair-psgg

[ECCV 2024 Oral] Code for our paper "A Fair Ranking and New Model for Panoptic Scene Graph Generation"
https://lorjul.github.io/fair-psgg/
10 stars 1 forks source link

Inquiry About SGG Evaluation Code #2

Closed slcheng97 closed 2 months ago

slcheng97 commented 2 months ago

Thank you for your inspiring work!

I attempted to reproduce your results in Table 1, but I could only locate the evaluation code for Predicate Classification. Unfortunately, I couldn’t find the evaluation code for the SGG (Scene Graph Generation) setting in this repository.

Could you please let me know where I can find the SGG evaluation code?

Thank you for your assistance!

lorjul commented 2 months ago

Thanks for your interest!

In principle the idea is as follows: use a segmentation model and generate a new PSG dataset (without the relations). Run inference using that dataset using DSFormer. Match the segmentation masks to the ground truth masks and calculate the metrics.

I will upload the related scripts, but it'll take some time to make them available to the public. I'm quite busy at the moment but will come back to you once I've uploaded the scripts.

lorjul commented 2 months ago

The evaluation scripts and instructions are now available.