model_compare_visual

Generate visual comparison artifacts between two model checkpoints.

Overview

model_compare_visual runs both models on a set of test images and produces side-by-side detection overlays for visual inspection of prediction differences.

Input Schema

Parameter	Type	Required	Default	Description
checkpoint_a	string	Yes	—	First model checkpoint.
checkpoint_b	string	Yes	—	Second model checkpoint.
image_paths	list[string]	Yes	—	Paths to test images.
output_dir	string	Yes	—	Directory for comparison outputs.
imgsz	integer	No	640	Inference image size.
conf	float	No	0.25	Confidence threshold.
device	string	No	"auto"	Inference device.

Output Schema

Field	Type	Description
checkpoint_a	string	First checkpoint path.
checkpoint_b	string	Second checkpoint path.
comparison_images	list[string]	Paths to generated comparison images.
image_count	integer	Number of comparisons generated.

Examples

CLI

curl -X POST http://127.0.0.1:7823/tools/model_compare_visual \
  -H "Authorization: Bearer $(cat ~/.fovux/auth.token)" \
  -H "Content-Type: application/json" \
  -d '{"checkpoint_a": "yolov8n.pt", "checkpoint_b": "yolov8s.pt", "image_paths": ["/data/img1.jpg"], "output_dir": "/tmp/compare"}'

Python

from fovux.tools.model_compare_visual import model_compare_visual

result = model_compare_visual(
    "yolov8n.pt", "yolov8s.pt", ["/data/img1.jpg"], output_dir="/tmp/compare"
)

Notes & Limits

• Output images are saved as PNG with side-by-side detection overlays.
• Large image sets may produce many output files.

Failure Modes

• Checkpoint resolution errors if either model is not found.
• File not found for any image path that does not exist.