Official implementation of ReconVLA: Reconstructive Vision-Language-Action Model as Effective Robot Perceiver.

PyTorch code for "Perceiver-VL: Efficient Vision-and-Language Modeling with Iterative Latent Attention" (WACV 2023)

AerithVLM is a hybrid remote sensing VLM combining DINOv3 and CLIP through a learned alignment head. Using a dual-encoder architecture, Vision Perceiver, and LLaMA backbone, it supports robust visual grounding, captioning, and open-ended geospatial reasoning.

A ROS2 Wrapper for DeepMind's Vision Perceiver IO Model

Hybrid Global Context Vision Transformer (GCViT) + Perceiver IO framework for medical image classification — Accepted in The Journal of Supercomputing.

A survey of Vision-Language Pre-training (VLP) focused on the BLIP lineage. Covers key architectural shifts including the Q-Former, Perceiver Resampler, and the integration of Diffusion Transformers with Autoregressive models.