Gradient Ascent for Caffe and Torch + Layer Viz #1009
Conversation
|
Layer Visualizations - Caffe |
|
Layer Visualizations - Torch (Notice Lack of Receptive Field Calculations) |
|
Gradient Ascent - Torch |
|
|
| """ | ||
| Get path to file containing model def for deploy/visualization | ||
| """ | ||
| return os.path.join(self.job_dir,"original.lua") |
There was a problem hiding this comment.
Maybe just call this network.lua for consistency? Or will there be more/'unoriginal' models?
There was a problem hiding this comment.
@TimZaman This PR is just of my top of branch before finishing my internship. Its the same as those two combines, but rebased with master, and Gradient Ascent for Torch.
There was a problem hiding this comment.
@TimZaman Yeah network.lua would definitely be a better name. Ill try updating this next time I get a chance :)
|
|
||
| fileReader.onreadystatechange = function (){ | ||
| if(fileReader.readyState === 4){ | ||
| if(fileReader.status === 200 || fileReader.readyState == 4){ |
There was a problem hiding this comment.
In line 50 you say
if(fileReader.readyState === 4){
and immediatelly after
if(fileReader.status === 200 || fileReader.readyState == 4){
So the latter is always true if the former is true. I bet you meant to say
(fileReader.status === 200 || fileReader.status == 0) here.
|
Closing for now (there are merge conflicts anyway). I'd like to revisit this. I've backed up this work at https://github.com/lukeyeager/DIGITS/tree/backup/lucas/gradientAscentWithLayerVis. |
|
@lukeyeager @TimZaman Sorry school has been super busy. But hopefully soon I'll get some time to revisit this and create some smaller pull requests (such as layer visualizations for Torch models, etc.) once I have a machine up and running. |
Do Not Merge!
Gradient Ascent For Torch and Caffe + Layer Visualizations.
Things left to do: