Amir Rosenfeld
Amir Rosenfeld
Hi, can you please add matconvnet to these benchmarks? Thanks, -Amir
The command `cat 20bn-something-something-v2-?? | tar zx` did not work for me directly. The 20 zip files must be first unzipped and only then can they be concatenated for tar...
When trying to train on the pascal dataset, it seems that a file "VOC/benchmark_RELEASEdataset/train_withValdata.txt" is missing. This is not a part of the original SBD distribution.
the function globexists is missing from rundissect_pytorch_external.sh: > ./script/rundissect_pytorch_external.sh: line 209: globexists: command not found it does exist, however, in rundissect.sh
Hi, Where in the code did you remove the mean and divide images by the std? I could not find this. If it's missing - this could be a potential...
Hi, cool framework! note that you add a layer of AvgPool2D with kernel=1 in the class VGG. This basically doesn't have any effect. Perhaps you meant AdaptiveAveragePool? In addition, the...
EDIT: Sorry - this is my mistake. I was confused because apparently in numpy.random.randint the "high" is exclusive, whereas the python random.randint isn't. You can delete this comment.
Very nice work! I suggest saving the pickle file for the dataset with protocol=2. This is faster in loading and saving and also uses 1.4GB instead of 5GB.
The variables steps, boxes, fm_sizes in net.py seem unused, though they're likely pretty important. Is this intentional? Possibly some other place does configure these values?
Hi, I suggest another missed reference: the paper "[priming neural networks](openaccess.thecvf.com/content_cvpr_2018_workshops/papers/w39/Rosenfeld_Priming_Neural_Networks_CVPR_2018_paper.pdf)" (shameless plug...) also deals with conditional computation, where object detection and semantic segmentation are shown to use side-info by...