Closed lyhangustc closed 6 years ago
The cropped images are from the ImageNet dataset. I am not sure if I am legally allowed to redistribute the cropped images from the ImageNet dataset.
I have consulted our company lawyer. I do not have the right to distribute the datasets.
Thanks for your reply.
Can you describe more details about these datasets so that I can redo the experiments myself? Like:
Sure.
I used VGG conv5 features for template matching. For each dog breed or cat specie, I manually cut 10-15 head portraits from the training set. They were my templates. I then performed multi-scale template matching, searching for the best candidate portrait (with most similar VGG conv5 feature) in each image. These matchers are then clustered into 100 clusters using their VGG conv5 features. We manually remove some clusters that contain no portraits. We also perform a similar step within each cluster.
Each category contains 1-10K unique images.
Yes. I mixed Egyptian cats, Persian cats, Tiger cats, and Tabby cats. I didn't try to translate the entire image since I wanted to get high resolution outputs for the animal face regions.
I am so interested in the Dog Breed Image Translation and the Cat Species Image Translation. Are the cropped cats and dogs datasets used in the experiments available?