search

UARK-AICV / AISFormer

[BMVC 2022] AISFormer: Amodal Instance Segmentation with Transformer
https://uark-aicv.github.io/AISFormer/
Apache License 2.0
28 stars 10 forks source link

Unable to reproduce the experimental results in the paper according to the default configuration file of the code. #5

Open mingheyuemankong opened 1 year ago

mingheyuemankong commented 1 year ago

I'm very sorry to bother you again! I think your work is excellent, but I want to seek the help of the author to solve my confusion. According to the configuration file in the code, I can't reproduce the results in the paper. The indexes such as AP on the three datasets are lower than the experimental data in the paper. For example, the AP error on D2SA and KINS datasets reaches 4%-6%, and the error on COCOA cls is even larger. What I want to ask are: (1) The weights in the three datasets configuration files under the configs folder are all WEIGHTS: "detectron2://ImageNetPretrained/MSRA/R-50.pkl" Can training with this weight achieve the data results in the paper? If so, why is there a big gap between my reproduced results and the paper? (2) If the author uses other weights to achieve such excellent data in the paper, can the author provide the weight link used? I look forward to the author's valuable reply. @trqminh

trqminh commented 1 year ago

Sorry for the late response, I see what is your concerning. I will update you as soon as I can.

mingheyuemankong commented 1 year ago

Yay, thanks and looking forward to your brilliant answers!

mingheyuemankong commented 1 year ago

Excuse me, is it still convenient to solve this problem about reproducing thesis data recently?

trqminh commented 1 year ago

I am working on it, hope you can wait a bit more

mingheyuemankong commented 1 year ago

Okay, thank you for your patient reply。

lcl-git-3d commented 12 months ago

Hi,have you solved this problem?

mingheyuemankong commented 12 months ago

No, I didn't solve the problem.