search

alexsax / midlevel-reps

Mid-Level Visual Representations Improve Generalization and Sample Efficiency for Learning Visuomotor Policies
http://perceptual.actor
MIT License
107 stars 19 forks source link

Semseg networks shouldn't apply tanh before softmax #11

Closed alexsax closed 4 years ago

alexsax commented 4 years ago

https://github.com/alexsax/midlevel-reps/blob/644bc7a30485841d7575c0b4967bb9d1dc13cf43/visualpriors/taskonomy_network.py#L190

To be fixed in future pull request: Remove this line if the network is semseg. Note: this doesn't affect the results in the paper since those used the latent representations rather than the output.

alexsax commented 4 years ago

fixed