Learning Abstract Visual Reasoning via Task Decomposition: A Case Study in Raven Progressive Matrices
One of the challenges in learning to perform abstract reasoning is that problems are often posed as monolithic tasks, with no intermediate subgoals.
Tags：Paper and LLMsVisual ReasoningPricing Type
- Pricing Type: Free
- Price Range Start($):
GitHub Link
The GitHub link is https://github.com/jakubkwiatkowski/abstract_compositional_transformer
Introduce
This repository, “jakubkwiatkowski/abstract_compositional_transformer,” contains an implementation of the Abstract Compositional Transformer (ACT) developed by the Neurosymbolic Systems Lab at Poznan University of Technology. The code includes models and utilities for property prediction and choice making. The usage instructions involve training the model in various phases using different tokenizers and masking techniques. Additionally, pre-trained weights are provided for evaluation.
One of the challenges in learning to perform abstract reasoning is that problems are often posed as monolithic tasks, with no intermediate subgoals.
Content
Code for models and utilities is available in repositories: You need pipenv to install the package. To install the package, run the following command: b. Train second phase – Task tokenizer with last masking
Related
We extend object tracking and 3D reconstruction algorithms to support continuous segmentation labels to leverage the advances in the 2D image segmentation, especially the Segment-Anything Model (SAM) which uses the pretrained neural network without additional training for new scenes, for 3D object segmentation.
