COMP558Project

This is a final course project of COMP 558 (Computer Vision).

Usage

MLV Implementation

The Bernhardt-Walther Lab at the University of Toronto created a MLV ToolBox GitHub repository. This toolbox provided us with the ability to analyze medial axes and perceptual organization cues. We have produced following test-files that used the codebase:

testing_LD_MAT.m - Creates line drawing & Computes MAT (outputting both images)

testing_Properties.m - Computes MAT feature scores and outputs all heat map visuals

testing_5050.m - Calls our created function new_drawSplitMATproperty to create contour splits of specified feature

new_drawSplitMATproperty - Created to divides pixels of an image into top 50% (high scores) and bottom 50% (low scores) based on the magnitude of the feature scores. This took the 5 images of each property (including our additional property) and produced 3 images for each (a total of 15): a line drawing with only the bottom 50% values visible (surrounded with a blue border), a line drawing with only the top 50% values visible (surrounded with a red border) and a line drawing with all values (bottom values colored blue, top values colored red. The function was modified to have an adjustable percentage to be more flexible (allow for different splits analysis)

Train a model

Example:

python main.py --pretrained [dataset-folder with train and val folders] --epochs 2 --batch-size 32 --lr 0.01

Test a model

Example:

python test.py --test_dir ../dataset/LD_dataset/val --model_path ./model_best.pth.tar --batch_size 128

Replace ../dataset/LD_dataset/val with the path to your test dataset.

Please follow the dataset structure, the label name should be the same as the folder name.

Dataset structure: dataset/ val/ class1/ image1.jpg image2.jpg ... class2/ image1.jpg image2.jpg ... ...

Reference:

1. MLV toolbox bwlabToronto/MLV_toolbox github.com
2. Contour/LineDrawing dataset given by Dirk Warther's lab https://osf.io/9squn/
3. Pytorch Example https://github.com/pytorch/examples/tree/main/imagenet