Recent years have witnessed a significant increase in the number of paper submissions to computer vision conferences. The sheer volume of paper submissions and the insufficient number of competent reviewers cause a considerable burden for the current peer review system. In this paper, we learn a classifier to predict whether a paper should be accepted or rejected based solely on the visual appearance of the paper (i.e., the gestalt of a paper). Experimental results show that our classifier can safely reject 50% of the bad papers while wrongly reject only 0.4% of the good papers, and thus dramatically reduce the workload of the reviewers. We also provide tools for providing suggestions to authors so that they can improve the gestalt of their papers.
Computer Vision Paper Gastalt dataset
- Low resolution (680 x 440) [link] (930 MB)
- High resolution (3400 x 2200) - Coming soon
Pre-trained weight for good/bad paper classifier [link] (44 MB)
- See Section 3.3 and 3.4 of the paper.
Pre-trained weight for random CVPR/ICCV paper generator [link] (270 MB)
- See Section 4.1: What does a good paper look like?
Pre-trained weight for bad-to-good paper generator [link] (44 MB)
- See Section 4.2: Learning bad-to-good paper translation
These instructions are reversed engineered, and hence not from the original authors.
TODO: include preprocessing starting from a pdf TODO: do inference
The pre-trained model runs as resnet18 in pytorch and require pypi's torch and torchvision. You also need to download the pre-trained weights.
# Install torch and torchvision
!pip3 install torch torchvision
# download weights
!wget http://filebox.ece.vt.edu/~jbhuang/project/gestalt/PaperNet.pth
Load the model using the downloaded PaperNet.pth file.
import torch
from torchvision.models import resnet18
papernet = resnet18(num_classes=2)
parameters = torch.load('PaperNet.pth')
papernet.load_state_dict(parameters)
For accepted vs. rejected classification use papernet.forward
papernet.forward # TODO: provide proper arguments
Stay tuned, not finished!