I get "all input arrays must have the same shape"... Why is that

[sean-wade]

I hope some one can reply me ... Thanks a lot!

[RogerChern]

Could you please describe which config you are using?

[sean-wade]

Could you please describe which config you are using?

I'm using this one: tridentnet_r101v2c4_c5_multiscale_addminival_3x_fp16.py

and it has started training for several hours... Maybe there is something wrong with my datasets?

[RogerChern]

Which dataset are you using and what about your hardware?

[sean-wade]

Which dataset are you using and what about your hardware?

It is my own dataset. Voc style.. But I think maybe some annotations has no objects. Will this cause this problem? My hardware is one Tesla P40

[RogerChern]

That should not be the problem. As the detection_train.py will filter the empty image. https://github.com/TuSimple/simpledet/blob/dd4eb159cf3934b949a89b745f00ddc62c0745c9/detection_train.py#L71

[RogerChern]

Could you please locate the input that will trigger this problem?

[sean-wade]

Could you please locate the input that will trigger this problem?

I add print function in the 'worker'. And I got this KeyError... I'm using 2 1080Ti GPU now... Thanks for replying...

[sean-wade]

I found that when using only 1 GPU, there is no problem... So I can only use 1 GPU now...

[conan1412]

@a954217436 i have the same question too,and only use 1 gpu can start the training,i can't find any clue to solve it