APS360 Project

Object Detection Process

Data Processing

• Check the original dimension of each frame. Can I reshape them to 448 by 448?
• If I reshape the pictures, the bounding box coordinates will need to be reshaped accordingly.
• We need to reformat the csv file to have appropriate labels. [is_person (optional), box_x, box_y, box_w, box_h]
• Need to have each frame as a training image and label. For each image, we need to have labels.
• Helper functions implementation.

Baseline

Performance:
Currently, we overfit on one camera angle cam_0.

• YOLOv1 loss
• YOLOv1 model
• YOLOv1 training

Primary Model (if time allows)

Performance:

• YOLOv3 loss
• YOLOv3 model
• YOLOv3 training

Person Re-identification Process:

Data Processing

• Reformat all images to the same size

Baseline

Performance:

• Simple script to compare the difference between 2 images
• Consider 2 images same if above a pre-set threshold.

Primary Model

Performance:

• In progress