Vijayabhaskar EV developed DINOv3 AutoBackbone support for the huggingface/transformers repository, expanding backbone options for computer vision models. He implemented new configuration, model, and backbone classes in Python, enabling multi-scale feature extraction and inter-layer access for detection and segmentation pipelines. His work ensured compatibility with Facebook’s get_intermediate_layers and improved normalization and hidden state handling, particularly when return_dict was set to False. Vijayabhaskar also contributed comprehensive tests and documentation, resolved merge conflicts, and fixed failing tests. This addition enhanced experimentation speed and stability, aligning backbone integration with Transformers principles and supporting broader deep learning and machine learning workflows.