Pretrained AdaPoinTr models

[mostafa501]

Hello @yuxumin. Thank you for your very good work. We are looking for pretrained models of AdaPoinTr so that we can obtain good results rather than using pretrained models of PoinTr, as in the attached image.

[yuxumin]

sorry, the best ckpt is lost due to server clean. But i find a ckpt for s55 (AdaPoinTr) in a previous issue. Have a try. https://github.com/yuxumin/PoinTr/issues/88#issuecomment-1383028718

[mostafa501]

Hi @yuxumin, I approciate your help, I need your help towards these results, I applied the pretrained models on ShapeNet55 benchmark,here, here, as you explained above. My data is about not completed chairs, as attached in "tested_files_and_results.zip" file, but the results are very bad as attached in images (red is input), (green is output). tested_files_and_results.zip

---

I used this snippet for run:"python tools/inference.py cfgs/ShapeNet55_models/AdaPoinTr.yaml ckpts/AdaPoinTr_s55.pth --pc_root demo2/ --save_vis_img --out_pc_root inference_result/" Can you please help me to complete my chairs data, tell me what should I do, knowing that I made normalization before network inference. Waiting your reply, thank you.

[liuyunxa]

你好，我也用了AdaPoinTr_s55.pth，我这里输出的文件和你的[tested_files_and_results.zip]对比，我这里没有pcd文件，这是什么原因呢？