search

rick-chang / OneNet

One Network to Solve Them All
GNU General Public License v3.0
85 stars 26 forks source link

Pre-trained model(Imagenet) results #1

Open rahulbaburaj opened 7 years ago

rahulbaburaj commented 7 years ago

I reproduced the result on the ring image using the pretrained model on imagenet. Since the image is 640 x 480, the last 32 pixels where left out. Still the image doesn't seem to match with result shown on the paper. cs_69999

Further the patches are very much visible. Do you do some kind of post processing after stitching the 64 x 64 images?

rahulbaburaj commented 7 years ago

Further on this particular image the results are poor. How should I change the alpha parameter if I'm using the pretrained model on imagenet? original_img parrot

githublzb commented 6 years ago

when I run paper_demo.py, there was an error: Traceback (most recent call last): File "/home/GithubDown/OneNet/admm/paper_demo.py", line 432, in testset_filelist = load_data.load_testset_path_list() File "../projector/load_celeb.py", line 74, in load_testset_path_list return load_testset_path()['image_path'].tolist() AttributeError: 'list' object has no attribute 'tolist' How do I overcome this? Thanks ahead.