PSMNet icon indicating copy to clipboard operation
PSMNet copied to clipboard

Error when running finetune.py

Open hubery4 opened this issue 4 years ago • 9 comments

Hi Jiaren,

When I use the pretrained model Scene_flow to finetune KITTI_2015, there is always a error: image Each time the index number is different. I just directly follow your instructions to train the dataset and I did not find the error point.

Could you give me some suggestions? Thank you so much.

hubery4 avatar Jun 20 '21 15:06 hubery4

hi,Have you solved this problem? I train with IndexError: index 99 is out of bounds for dimension 1 with size 1 image

Jack47746 avatar Jun 25 '21 13:06 Jack47746

It seems the problem of the dimension of 'pred_disp'. 'true_disp' has 3 dimension while 'pred_disp' has 4. You need add ':' in pred_disp. image

hubery4 avatar Jun 25 '21 14:06 hubery4

thank you so much!Have you encountered such a problem? Why did I stop running halfway through the run, and there was no error. image

Jack47746 avatar Jun 26 '21 04:06 Jack47746

@Jack47746 I have also encountered this problem both in Colab and my own computer. It seems the RAM memory problem. It needs much more memory (>32GB). But I have not validated it. If you have some methods, please share it with me. Thank you.

hubery4 avatar Jun 26 '21 09:06 hubery4

change the batchsize? My batchsize is 3.Do you speak chinese, do you have QQ or Wechat?

Jack47746 avatar Jun 26 '21 10:06 Jack47746

@Jack47746 Changing batchsize seems not working. I have WeChat. You could directly email me ([email protected]) to give me the wechat number and I can add your wechat.

hubery4 avatar Jun 26 '21 10:06 hubery4

https://github.com/JiaRenChang/PSMNet/issues/226

michaelnguyen11 avatar Oct 24 '21 17:10 michaelnguyen11

@Jack47746 Changing batchsize seems not working. I have WeChat. You could directly email me ([email protected]) to give me the wechat number and I can add your wechat.

你好,我加个分号但是又报cpu占用不够了,请问你知道方法吗?我微信xue666630

yuankang1234 avatar Oct 10 '22 08:10 yuankang1234