Skip to content

README.md

Latest commit

 

History

History
139 lines (101 loc) · 3.05 KB

README.md

File metadata and controls

139 lines (101 loc) · 3.05 KB

Visualization

Scripts for running each network on a single image or a video and rendering the predictions as overlays. Each network has its own subfolder with image_visualization.py and video_visualization.py.

All scripts use the inference classes from inference/ internally.

FreespaceSeg — Drivable Area Segmentation

Single Image

python FreespaceSeg/image_visualization.py \
  -p /path/to/checkpoint.pth \
  -i /path/to/image.jpg

Displays the input image blended (50% alpha) with the 2-class prediction mask:

Class Color
Non-drivable Blue (255, 93, 61)
Drivable Cyan (220, 255, 0)

Video

python FreespaceSeg/video_visualization.py \
  -p /path/to/checkpoint.pth \
  -i /path/to/input.mp4 \
  -o output_freespaceseg \
  [-v]
Argument Description
-o Output base name — saves {name}.avi (MJPG codec)
-v Show live preview window

FreespaceContour — Drivable Boundary Contour

Single Image

python FreespaceContour/image_visualization.py \
  -p /path/to/checkpoint.pth \
  -i /path/to/image.jpg

Draws 37 red circles connected by lines tracing the predicted drivable boundary. Each circle is placed at a distance (bin index × 10 px) along its ray, sweeping left to right across the bottom of the image.

Video

python FreespaceContour/video_visualization.py \
  -p /path/to/checkpoint.pth \
  -i /path/to/input.mp4 \
  -o output_contour.mp4

Output is saved as an MP4 at the original video resolution.

Elevation — Terrain Elevation Map

Requires a camera parameters file (flattened 3×4 [R|t], .npy, shape (12,)).

Single Image

python Elevation/image_visualization.py \
  -p /path/to/checkpoint.pth \
  -i /path/to/image.jpg \
  -c /path/to/camera_params.npy \
  [--alpha 0.5]

Displays a side-by-side view:

  • Left: input image blended with a jet-colormap elevation overlay
  • Right: color bar with elevation scale from −0.5 m (bottom) to +1.5 m (top)

The elevation map is predicted at H/4 × W/4 (80×160 for a 320×640 input) and upsampled back to the original image size for display.

Video

python Elevation/video_visualization.py \
  -p /path/to/checkpoint.pth \
  -i /path/to/input.mp4 \
  -c /path/to/camera_params.npy \
  -o output_elevation.mp4

AutoSpeed — Speed Limit / Object Detection

Single Image

python AutoSpeed/image_visualization.py \
  -p /path/to/best.pt \
  -i /path/to/image.jpg

Draws colored bounding boxes for each detection:

Class ID Color
1 Red
2 Yellow
3 Cyan

Video

python AutoSpeed/video_visualization.py \
  -p /path/to/best.pt \        # or /path/to/model.onnx
  -i /path/to/input.mp4 \
  -o output_autospeed \
  [-v]

Automatically detects whether the checkpoint is a PyTorch (.pt) or ONNX (.onnx) model and selects the appropriate inference backend.

Argument Description
-o Output base name — saves {name}.avi
-v Show live preview at 960 px width