zero-shot capabilities
The ability of a model to perform a task it has not been explicitly trained on, often relying on generalization from related tasks or rich contextual understanding.
- Advancing Compositional Awareness in CLIP with Efficient Fine-Tuning
- CLIPTTA: Robust Contrastive Vision-Language Test-Time Adaptation
- Diffusion Transformers as Open-World Spatiotemporal Foundation Models
- Native-Resolution Image Synthesis
- OpenWorldSAM: Extending SAM2 for Universal Image Segmentation with Language Prompts
- Towards 3D Objectness Learning in an Open World