Which model ckpt to use for novel affordance detection of in-the-wild images

[tumble3eed]

Thanks for your great work!
I am trying to use your model in my task and need to perform new affordance segmentation on in-the-wild object images.
I need a model that has seen more data and is more accurate. May I ask if your seen and unseen checkpoints were trained on the seen and unseen split of AGD20K, respectively? Which checkpoint should I use? Thank you!

[Reagan1311]

Hi, there. 1) For your specific task, you may need to re-annotate one-shot data, as the annotation standard of AGD20K might differ from what you need. 2) Yes, the seen and unseen checkpoints were trained on their respective splits of AGD20K. However, these checkpoints are not specifically tailored for novel affordance segmentation. Therefore, we recommend modifying the text prompt learning to train a new model as suggested in #5.