Closed lxmaczj closed 1 month ago
your results?
ye,I followed the code provided to reproduce and the results on the OLMdataset were very different from those in the paper。For example unseen_occ_acc:
pretrained_split2 {'seen_err': 4.020338509585563, 'seen_acc': 0.993883792048929, 'seen_occ_err': 9.4861058260868, 'seen_occ_acc': 0.8659242761692652, 'unseen_err': 2.491533233652202, 'unseen_acc': 0.9948707427164555, 'unseen_occ_err': 10.088256396094332, 'unseen_occ_acc': 0.8020767111676205}
ours {'seen_err': 3.894989475453137, 'seen_acc': 0.9938837920489294, 'seen_occ_err': 11.710244415221577, 'seen_occ_acc': 0.8149220489977737, 'unseen_err': 3.3169533221516776, 'unseen_acc': 0.9725071809601985, 'unseen_occ_err': 13.278622619779107, 'unseen_occ_acc': 0.6916719643992372}
------------------ 原始邮件 ------------------ 发件人: "Tianfu18/diff-feats-pose" @.>; 发送时间: 2024年9月19日(星期四) 下午3:18 @.>; @.**@.>; 主题: Re: [Tianfu18/diff-feats-pose] OLMresults (Issue #5)
your results?
— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @.***>
The results are as expected, with the instability on the O-LM dataset caused by randomness. Simply train multiple times and select the right checkpoint to reproduce the results. Often, the best checkpoint is found in the first few epochs, not necessarily the last one.
Thank you very much for your answer.
------------------ 原始邮件 ------------------ 发件人: "Tianfu18/diff-feats-pose" @.>; 发送时间: 2024年9月19日(星期四) 下午3:48 @.>; @.**@.>; 主题: Re: [Tianfu18/diff-feats-pose] OLMresults (Issue #5)
The results are as expected, with the instability on the O-LM dataset caused by randomness. Simply train multiple times and select the right checkpoint to reproduce the results. Often, the best checkpoint is found in the first few epochs, not necessarily the last one.
— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @.***>
On OLM data set,The result of training and then testing is quite different from the result of the given pretrained test.