The Spatial Blindspot of Vision-Language Models

Vision-language models (VLMs) have advanced rapidly, but their ability to capture spatial relationships remains a blindspot. Current VLMs are typically built with contrastive language-image pretraining (CLIP) style image encoders. The training recipe often flattens images into 1D patch sequences, discarding the 2D structure necessary for spatial reasoning. We argue that this lack of spatial awareness is a missing dimension in VLM design and a bottleneck for applications requiring spatial grounding, such as robotics and embodied AI. To address this, we investigate (i) image encoders trained with alternative objectives and (ii) 2D positional encodings. Our experiments show that these architectural choices can lead to improved spatial reasoning on several benchmarks.

Histo-fetch -- On-the-fly processing of gigapixel whole slide images simplifies and speeds neural network training

We created a custom pipeline (histo-fetch) to efficiently extract random patches and labels from pathology whole slide images (WSIs) for input to a neural network on-the-fly. We prefetch these patches as needed during network training, avoiding the need for WSI preparation such as chopping/tiling. We demonstrate the utility of this pipeline to perform artificial stain transfer and image generation using the popular networks CycleGAN and ProGAN, respectively.