https://github.com/georgepearse/visionchain

Framework to support stitching together multiple models to create effective Computer Vision systems, as accurate or fast as you need them to be.
https://github.com/georgepearse/visionchain

active-learning computer-vision heuristics object-detection

Last synced: 10 months ago
JSON representation

Framework to support stitching together multiple models to create effective Computer Vision systems, as accurate or fast as you need them to be.

• Host: GitHub
• URL: https://github.com/georgepearse/visionchain
• Owner: GeorgePearse
• License: apache-2.0
• Created: 2023-06-11T20:44:22.000Z (almost 3 years ago)
• Default Branch: main
• Last Pushed: 2025-06-18T20:26:40.000Z (10 months ago)
• Last Synced: 2025-06-19T04:06:12.339Z (10 months ago)
• Topics: active-learning, computer-vision, heuristics, object-detection
• Language: Python
• Homepage: https://georgepearse.github.io/VisionChain/
• Size: 49.9 MB
• Stars: 5
• Watchers: 1
• Forks: 1
• Open Issues: 2
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

          
# VisionChain

Framework to support common preprocessing and postprocessing steps, along with computer vision heuristics built with opencv, and voting systems ontop of those heuristic-model combos. 

Could add a hyperparameter optimizer for things like setting the confidence at which to intervene.

Constraining artificial stupidity and giving AI some hand rails.

To Do: 

- [ ] Create some templates, e.g. cascade model

- [ ] Support VQA models on crops or similar

- [ ] Support NN Classifiers based on the embeddings of multiple models.

- [ ] Support Segmentation models

- [ ] Support more from https://github.com/IDEA-Research/Grounded-Segment-Anything/tree/main

- [ ] Integrate TaskMatrix a bit https://github.com/microsoft/TaskMatrix

```python

model = vc.ModelChain(

    [

        vc.ConditionalDetector(

            model=yolov8,

        ),

        vc.ConditionalDetector(

            model=grounded_sam, 

            frame_level_condition = lambda predictions: any([score < 0.5 for score in predictions.scores]),

            prediction_level_condition = lambda pred: 'cat' in pred.label,

        ),

        vc.ConditionalClassifier(

            model=nn_classifier,

            prediction_level_condition = lambda pred: 'dog' in pred.label,

        ),

    ],

    log_level = 'verbose',

)

```