The images to be classified must be in a subdirectory that is alone in its parent directory, e.g. /path/to/directory/classify/<pictures>
Then, the option -classify should be invoked like this -classify /path/to/directory/
I added a classifyLoader that is derived from testLoader but only does forward passes and stores the top-1 predictions for each classes into a file called "predictions.txt".
I also added a Throughput measurement in data.lua that helps finding out the right batchSize.
not really happy with the if test
if opt.classify == '' then -- no classification
if you know a cleaner way...
Also, perhaps the option -retrainshould be renamed to -modelCheckpoint or something...
The images to be classified must be in a subdirectory that is alone in its parent directory, e.g. /path/to/directory/classify/
<pictures
> Then, the option -classify should be invoked like this -classify /path/to/directory/I added a classifyLoader that is derived from testLoader but only does forward passes and stores the top-1 predictions for each classes into a file called "predictions.txt".
I also added a Throughput measurement in data.lua that helps finding out the right batchSize.