Skip to content

Latest commit

 

History

History
51 lines (42 loc) · 2.04 KB

README.md

File metadata and controls

51 lines (42 loc) · 2.04 KB
Raw

GigaVision

GigaVision is a program that seeks to revolutionize computer vision when it meets gigapixel videography with both wide field-of-view and high-resolution details. The project contains visualization notebooks that helps understanding the diversity of gigapixel detection and tracking dataset. It also includes a search space which explore best possible image size with respect to best accuracy and inference time.

Notebooks

A Jupyter Notebook for visualizing annotations with added features for a better user experience.

A Jupyter Notebook for generating COCO annotations from JSON files.

Plot search space results in 3D.

Visualize bounding box distribution of each tracking ID in a particular sequence.

Getting Started

Prerequisites

  • Jupyter Notebook
  • Matplotlib
  • NumPy
  • OpenCV
  • PyCOCO Tools

Explore Search Space

Please install sahi

python gv/explore.py \
    --conf 0.3 \
    --height_ratio 0.15 \
    --width_ratio 0.05 \
    --IOS_thresh 0.7 \
    --model_path yolov7-e6e.pt \
    --model_type yolov7 \
    --path val/ \
    --annotations_path src/val.json \
    --csv src/training_data.csv \
    --outfile results.json
  • --conf --> Confidence threshold
  • --height_ratio --> Overlap height ratio
  • --width_ratio --> Overlap width ratio
  • --IOS_thresh --> Intersection over union threshold
  • --model_path --> path to model weights
  • --model_type --> type of model (yolov7, detectron2, mmdet)
  • --path --> Path to the image folder
  • --annotations_path --> Path to the annotations file
  • --csv --> Path to the training_data.csv file