Visual Programming: Compositional visual reasoning without training
This work addresses the challenge of expanding AI systems to handle a wide range of complex visual tasks for users, though it is incremental as it builds on existing neuro-symbolic and in-context learning methods.
The paper tackles the problem of performing complex and compositional visual reasoning tasks without task-specific training by introducing VISPROG, a neuro-symbolic approach that uses large language models to generate modular programs executed with off-the-shelf vision models, achieving flexibility across four diverse tasks such as visual question answering and image editing.
We present VISPROG, a neuro-symbolic approach to solving complex and compositional visual tasks given natural language instructions. VISPROG avoids the need for any task-specific training. Instead, it uses the in-context learning ability of large language models to generate python-like modular programs, which are then executed to get both the solution and a comprehensive and interpretable rationale. Each line of the generated program may invoke one of several off-the-shelf computer vision models, image processing routines, or python functions to produce intermediate outputs that may be consumed by subsequent parts of the program. We demonstrate the flexibility of VISPROG on 4 diverse tasks - compositional visual question answering, zero-shot reasoning on image pairs, factual knowledge object tagging, and language-guided image editing. We believe neuro-symbolic approaches like VISPROG are an exciting avenue to easily and effectively expand the scope of AI systems to serve the long tail of complex tasks that people may wish to perform.