Testing

Planova's pipeline has end-to-end tests that exercise the full floor-plan-to-scene conversion flow using a real floor plan image and a mock VLM response.

Test File

All pipeline E2E tests live in:

src-tauri/src/pipeline/test_e2e.rs

Running Tests

From the src-tauri/ directory:

cd src-tauri
cargo test --lib test_pipeline_e2e -- --nocapture

The --nocapture flag prints detailed diagnostic output (room areas, alignment scores, segment counts) to stdout during the run.

To run all library tests:

cargo test --lib

Test Cases

1. test_wall_mask_extraction

Tests the wall mask extraction step in isolation.

What it does:

1. Preprocesses the test image (assets/plane-design-3.png).
2. Extracts a binary wall mask via wall_mask::extract_wall_mask.
3. Computes the ratio of wall pixels to total image pixels.

Assertions:

Check	Condition
Wall pixel ratio	Between 0.5% and 20%
Title block region (y < 150)	Fewer than 500 wall pixels
Legend region (y > 900)	Fewer than 500 wall pixels

These checks verify that the mask correctly identifies structural walls while ignoring annotations, title blocks, and legends.

2. test_wall_graph_segments

Tests wall skeleton graph construction from the wall mask.

What it does:

1. Runs preprocessing and wall mask extraction.
2. Builds the wall skeleton graph via wall_graph::build_wall_graph.
3. Inspects the resulting segments and junctions.

Assertions:

Check	Condition
Segment count	Between 3 and 80
No full-span segments	No horizontal segment exceeds 95% of image width; no vertical segment exceeds 95% of image height
No title-block segments	No segment midpoint with y < 150
No legend segments	No segment midpoint with y > 900

3. test_full_pipeline_with_mock_vlm

Tests the complete pipeline from preprocessing through scene validation.

What it does:

1. Preprocesses the image and extracts wall mask and wall graph.
2. Constructs a mock VLM response (see below).
3. Builds the PlanGraph from wall segments + VLM response.
4. Converts PlanGraph to a HomeSceneJSON scene.
5. Runs scene repair (repair::repair_scene).
6. Computes wall alignment metrics (alignment::compute_alignment).
7. Validates the final scene (validate::validate_scene_with_alignment).

Assertions:

Check	Condition
Face count	At least 2 faces
Room count	At least 2 rooms
Alignment overall score	Greater than 0.5
Walls with room_refs	At least 1
Scale mpp (best candidate)	Between 0.005 and 0.015
Per-room max dimension	Less than 15 m
Total area	Between 20 m² and 100 m²

Mock VLM Response Structure

The mock response simulates what the VLM returns when given a floor plan image:

{
  "detected_rooms": [
    {
      "type": "living_room",
      "name": "客厅",
      "centroid": [300, 500],
      "confidence": 0.9
    },
    {
      "type": "bedroom",
      "name": "卧室",
      "centroid": [900, 400],
      "confidence": 0.9
    },
    {
      "type": "bathroom",
      "name": "卫生间",
      "centroid": [900, 700],
      "confidence": 0.9
    }
  ],
  "detected_doors": [
    {
      "position": [700, 400],
      "width_meters": 0.9,
      "connected_rooms": ["living_room", "bedroom"],
      "swing_direction": "right_inward",
      "confidence": 0.8
    },
    {
      "position": [700, 700],
      "width_meters": 0.9,
      "connected_rooms": ["living_room", "bathroom"],
      "swing_direction": "right_inward",
      "confidence": 0.8
    }
  ],
  "detected_windows": [
    {
      "position": [300, 200],
      "width_meters": 1.2,
      "wall_side": "north",
      "confidence": 0.7
    },
    {
      "position": [900, 200],
      "width_meters": 1.2,
      "wall_side": "north",
      "confidence": 0.7
    }
  ],
  "scale_info": {
    "detected": true,
    "meters_per_pixel": 0.01
  },
  "dimension_annotations": [
    { "text": "8400", "position": [600, 100], "direction": "horizontal" },
    { "text": "6000", "position": [100, 500], "direction": "vertical" }
  ]
}

Key fields:

• detected_rooms -- Room labels with centroid pixel positions and type.
• detected_doors -- Door positions, widths in meters, connected rooms, and swing direction.
• detected_windows -- Window positions, widths, and wall side.
• scale_info -- Initial scale estimate (meters per pixel).
• dimension_annotations -- OCR-detected dimension labels from the floor plan image (values in millimeters).

Test Artifacts

All test outputs are saved to /tmp/planova_test/ (or /tmp/planova_test_full/ for the full pipeline test). This directory includes:

• preprocessed.png -- Preprocessed floor plan image
• wall_mask.png -- Binary wall mask
• wall_graph.json -- Wall skeleton graph data
• overlay_alignment.png -- Alignment overlay visualization

The directory is created automatically before each test run.