Search

Skin cancer is a possible curable disease when detected in the early stage, but skin cancer diagnosis is difficult for people in developing countries where residents lack access to proper healthcare. In this paper, we present a low-cost, easy-to-use, and internet-free prescreening solution to detect cancer earlier in rural areas where medical resources are scarce. We deliver a prototype of a device that can classify the skin anomaly into seven major categories and calculate the area segmentation. The prototype we designed includes a Raspberry Pi 3B+, Pi camera, magnifying camera attachment, a convolution neural network powering skin cancer recognition, another network for skin cancer boundary segmentation, and an interactive touchscreen user interface in a custom enclosure. We trained a MobileNetV2 for skin cancer recognition and a U-Net for skin cancer boundary segmentation on the Skin Cancer MNIST dataset collected by The International Skin Imaging Collaboration and used it as our skin cancer recognition model. We then deployed both models onto a Raspberry Pi, and made it into a handy device that takes a close-up picture and prescreen the patient's skin quickly. Through this research process, we found out that it is possible to run heavy computation such as deep learning on a Raspberry Pi and be packaged into a low cost handheld device for screening purposes for a very low cost.

In this paper, we investigate the use of an unsupervised label clustering technique and demonstrate that it enables substantial improvements in visual relationship prediction accuracy on the Person in Context (PIC) dataset. We propose to group object labels with similar patterns of relationship distribution in the dataset into fewer categories. Label clustering not only mitigates both the large classification space and class imbalance issues, but also potentially increases data samples for each clustered category. We further propose to incorporate depth information as an additional feature into the instance segmentation model. The additional depth prediction path supplements the relationship prediction model in a way that bounding boxes or segmentation masks are unable to deliver. We have rigorously evaluated the proposed techniques and performed various ablation analysis to validate the benefits of them.