3D scene style transfer with 2D style image by differential rendering:

• Internship performance exceeds mentor/peers' expectation in review.
• Learned style transfer, 3D mesh and rendering from scratch in one week.
• Utilized PyTorch3D & nvdiffrast as differential rendering to generate 2D views.
• Optimized texture maps by style transfer between 2D rendered images and style image.
• Preserved object style consistency by semantic style transfer.

Symmetric learning for Fracture Detection in Pelvic Trauma X-ray:

• Paper accepted by ECCV 2020 with poster presentation.
• Mimicked radiologists' practice by comparing vertical asymmetric areas via Siamese network.
• Aligned Siamese features according to GNN-detected pelvic structure landmarks.
• Learned anatomical asymmetry explicitly by novel pixel-wise contrastive loss.

Deep Hierarchical Multi-label Classification of Chest X-ray:

• Paper accepted by MIDL 2019 with oral presentation.
• Special invitation to Journal ``Medical Image Analysis" and paper accepted.
• Followed clinical taxonomy to construct hierarchical multi-label classification.
• Developed a two-stage training procedure to fit the extreme label imbalance dataset.
• Derived a numerically stable math formulation to avoid floating point underflow calculating loss.

Lung nodule detection in CT images:

• Achieved rank 6 out of 2887 teams in the Skylake competition sponsored by Intel and Alibaba.
• Applied PyTorch, 3D UNet and Caffe, Faster RCNN to detect lung nodules in 1000 CT scans.
• Used fusion method to achieve false positive reduction.