albumentation

Cross-library augmentation toolbox supporting 300 operators over 8 libraries + AI transforms

Salt Body Segmentation in Seismic Imaging using Semi-Supervised Learning

Developed a customized CNN-based model (AOD) for small aerial object detection in images of the sky captured from ground level. Enhanced detection accuracy for extremely small-scale objects like drones, birds, and aircraft by optimizing backbone architecture, applying targeted augmentations, and refining anchor box settings.

PyTorch DeepLabV3/V3+ semantic segmentation on Cityscapes: training, mIoU evaluation, inference overlays, and submission-ready labelIds export.