search

mapbox / robosat

Semantic segmentation on aerial and satellite imagery. Extracts features such as: buildings, parking lots, roads, water, clouds
MIT License
2.02k stars 383 forks source link

Save loss and IoU numbers as text file #92

Closed maning closed 6 years ago

maning commented 6 years ago

Currently rs train saves only history plot and checkpoint pth. For very large epoch numbers, its hard to understand from the saved plot the small variations in the model after every epoch. Saving the result as a text file would help dig into the numbers better.

daniel-j-h commented 6 years ago

I agree. We could think about integrating more full blown solutions like tensorboard or visdom. But that may be too much for the user-base we are targeting here and the use-cases we have in mind.

Maybe writing out a simple text file you can easily evaluate would be a good idea.

Here are the related code locations where we could add it:

https://github.com/mapbox/robosat/blob/6c7f547d2dd303f8a68b4fbd2ba60583348d7816/robosat/tools/train.py#L77-L98

daniel-j-h commented 6 years ago

https://github.com/mapbox/robosat/pull/118 saves a log file. Closing this.