Runtime error when training imitation learning model

duckietown / gym-duckietown

Self-driving car simulator for the Duckietown universe

http://duckietown.org

Other

52 stars 19 forks source link

Runtime error when training imitation learning model #125

Closed duckieT closed 5 years ago

duckieT commented 5 years ago

After running experiments/train_imitation.py --map-name loop_obstacles, I got this error:

RuntimeError: size mismatch, m1: [2 x 62016], m2: [2816 x 256 ] at /pytorch/aten/src/THC/generic/THCTensorMathBlas.cu:249

Can anyone please help :(

bhairavmehta95 commented 5 years ago

So the train_imitation code is probably quite old, in which case the convolution filters will not produce the same size output as they used to (We changed the image size that comes out of the simulator).

You can take a look at the model getting called and edit the numbers explictly (it should be in the first fully connected layer). I will have some time early next week to fix it otherwise.

bhairavmehta95 commented 5 years ago

Follow in #124 and #130