To gain some perspective, try farm5.static.flickr.com/4096/492138304 on cloudcv.org/vqa/ and ask almost anything. #dlearn
Conversation
Replying to
I can't explain why it recognizes so few people, especially with those mirrors
2
Replying to
would be neat if it presented evidence for its decisions, wouldn’t it?
1
Replying to
as in "these blobs I consider X"? or something deeper?
Replying to
by tendency hard w/ neural networks but the convolutional part of the model has features that could be visualized
1
Replying to
I pass, I would need more than my superficial knowledge of convnets here :/
2
Show replies