model-compare

Name: model-compare
Author: rawwerks

98stars11forks

How to install model-compare?

rawwerks/vibecad/model-compare

$npx -y skills add rawwerks/VibeCAD --skill model-compare

Installs into the current project.

Summary

Compare 3D CAD models using boolean operations (IoU, Dice, precision/recall). Use when evaluating generated models against gold references, diffing CAD revisions, or computing similarity metrics for ML training. Triggers on: model diff, compare models, IoU, intersection over unio

Files · 1

View on GitHub

SKILL.md

1	# 3D Model Comparison Tool
2
3	Compare CAD models using boolean operations to compute similarity metrics like IoU, Dice, precision, and recall. Useful for:
4
5	- Evaluating ML-generated models against gold references
6	- Comparing revisions of CAD designs
7	- Computing metrics for training 3D generative models
8	- Visualizing geometric differences
9
10	## Quick Start
11
12	```bash
13	# Compare two STEP files
14	uvx --from build123d python scripts/model_diff.py reference.step generated.step
15
16	# JSON output for training pipelines
17	uvx --from build123d python scripts/model_diff.py ref.step gen.step --json --no-export
18
19	# Demo mode (no files needed)
20	uvx --from build123d python scripts/model_diff.py --demo
21	```
22
23	## Supported Formats
24
25	\| Format \| Extension \| Notes \|
26	\|--------\|-----------\|-------\|
27	\| STEP \| `.step`, `.stp` \| Recommended - full CAD fidelity \|
28	\| BREP \| `.brep` \| OpenCASCADE native format \|
29	\| STL \| `.stl` \| Mesh format - may have boolean issues \|
30
31	## Output Metrics
32
33	### Primary Metrics (for ML training)
34
35	\| Metric \| Range \| Description \|
36	\|--------\|-------\|-------------\|
37	\| IoU (Jaccard) \| 0-1 \| `\|A∩B\| / \|A∪B\|` - standard similarity \|
38	\| Dice (F1) \| 0-1 \| `2\|A∩B\| / (\|A\|+\|B\|)` - more sensitive to small overlaps \|
39	\| Precision \| 0-1 \| `\|A∩B\| / \|B\|` - how much of generated is correct \|
40	\| Recall \| 0-1 \| `\|A∩B\| / \|A\|` - how much of reference was captured \|
41
42	### Diagnostic Metrics
43
44	\| Metric \| Description \|
45	\|--------\|-------------\|
46	\| `volume_ratio` \| B/A volume ratio (1.0 = same size) \|
47	\| `center_offset` \| Distance between centers of mass \|
48	\| `bbox_iou` \| Bounding box IoU (coarse alignment) \|
49	\| `size_ratio_x/y/z` \| Per-axis scale comparison \|
50	\| `surface_ratio` \| Surface area comparison \|
51
52	### Interpretation
53
54	The tool provides automatic interpretation:
55	- Over-generating: Low precision, high extra geometry
56	- Under-generating: Low recall, missing geometry
57	- Size issues: Volume ratio far from 1.0
58	- Position issues: Large center offset
59
60	## CLI Options
61
62	```
63	usage: model_diff.py [-h] [-o OUTPUT_DIR] [--json] [--no-export] [--demo]
64	[reference] [generated]
65
66	positional arguments:
67	reference Reference/gold model file (STEP, BREP, or STL)
68	generated Generated/predicted model file to compare
69
70	options:
71	-o, --output-dir Output directory for GLB files (default: .)
72	--json Output only JSON metrics (for pipelines)
73	--no-export Skip exporting GLB visualization files
74	--demo Run with built-in demo models
75	```
76
77	## Output Files
78
79	When `--no-export` is not set, produces GLB files for visualization:
80
81	\| File \| Description \|
82	\|------\|-------------\|
83	\| `diff_reference.glb` \| The reference model (A) \|
84	\| `diff_generated.glb` \| The generated model (B) \|
85	\| `diff_missing.glb` \| Geometry in A but not B (under-generation) \|
86	\| `diff_extra.glb` \| Geometry in B but not A (over-generation) \|
87	\| `diff_common.glb` \| Geometry in both (correct match) \|
88
89	## Example: Training Pipeline Integration
90
91	```bash
92	# Batch evaluation
93	for gen in outputs/*.step; do
94	uvx --from build123d python model_diff.py gold.step "$gen" --json --no-export
95	done \| jq -s '{
96	avg_iou: (map(.iou) \| add / length),
97	avg_precision: (map(.precision) \| add / length),
98	avg_recall: (map(.recall) \| add / length)
99	}'
100	```
101
102	## Example: Loss Function
103
104	```python
105	# In your training code, use metrics for loss:
106	loss = (
107	(1 - metrics['iou']) * 1.0 + # Primary shape match
108	abs(1 - metrics['volume_ratio']) * 0.5 + # Scale accuracy
109	metrics['center_offset'] * 0.1 # Position accuracy
110	)
111	```
112
113	## How It Works
114
115	The tool uses boolean operations from OpenCASCADE (via build123d):
116
117	```
118	Missing = Reference - Generated (A - B)
119	Extra = Generated - Reference (B - A)
120	Common = Reference & Generated (A ∩ B)
121	Union = Reference + Generated (A ∪ B)
122
123	IoU = volume(Common) / volume(Union)
124	Dice = 2 * volume(Common) / (volume(A) + volume(B))
125	Precision = volume(Common) / volume(B)
126	Recall = volume(Common) / volume(A)
127	```
128
129	## Sample Output
130
131	```
132	=================================================================
133	3D MODEL COMPARISON REPORT
134	Reference (A) vs Generated (B)
135	=================================================================
136
137	──────────────────────────── VOLUMES ────────────────────────────
138	Reference (A): 51,433.629
139	Generated (B): 45,904.426
140	Intersection (A∩B): 42,292.031
141	Missing (A-B): 9,141.598 (17.8% of A)
142	Extra (B-A): 3,612.395 (7.9% of B)
143
144	──────────────────────── PRIMARY METRICS ───────────────

People who liked this also viewed

6 picks

Type

model-compare

98stars11forks

View on GitHub

How to install model-compare?

rawwerks/vibecad/model-compare

$npx -y skills add rawwerks/VibeCAD --skill model-compare

Installs into the current project.

Summary

Files · 1

View on GitHub

SKILL.md

1	# 3D Model Comparison Tool
2
3	Compare CAD models using boolean operations to compute similarity metrics like IoU, Dice, precision, and recall. Useful for:
4
5	- Evaluating ML-generated models against gold references
6	- Comparing revisions of CAD designs
7	- Computing metrics for training 3D generative models
8	- Visualizing geometric differences
9
10	## Quick Start
11
12	```bash
13	# Compare two STEP files
14	uvx --from build123d python scripts/model_diff.py reference.step generated.step
15
16	# JSON output for training pipelines
17	uvx --from build123d python scripts/model_diff.py ref.step gen.step --json --no-export
18
19	# Demo mode (no files needed)
20	uvx --from build123d python scripts/model_diff.py --demo
21	```
22
23	## Supported Formats
24
25	\| Format \| Extension \| Notes \|
26	\|--------\|-----------\|-------\|
27	\| STEP \| `.step`, `.stp` \| Recommended - full CAD fidelity \|
28	\| BREP \| `.brep` \| OpenCASCADE native format \|
29	\| STL \| `.stl` \| Mesh format - may have boolean issues \|
30
31	## Output Metrics
32
33	### Primary Metrics (for ML training)
34
35	\| Metric \| Range \| Description \|
36	\|--------\|-------\|-------------\|
37	\| IoU (Jaccard) \| 0-1 \| `\|A∩B\| / \|A∪B\|` - standard similarity \|
38	\| Dice (F1) \| 0-1 \| `2\|A∩B\| / (\|A\|+\|B\|)` - more sensitive to small overlaps \|
39	\| Precision \| 0-1 \| `\|A∩B\| / \|B\|` - how much of generated is correct \|
40	\| Recall \| 0-1 \| `\|A∩B\| / \|A\|` - how much of reference was captured \|
41
42	### Diagnostic Metrics
43
44	\| Metric \| Description \|
45	\|--------\|-------------\|
46	\| `volume_ratio` \| B/A volume ratio (1.0 = same size) \|
47	\| `center_offset` \| Distance between centers of mass \|
48	\| `bbox_iou` \| Bounding box IoU (coarse alignment) \|
49	\| `size_ratio_x/y/z` \| Per-axis scale comparison \|
50	\| `surface_ratio` \| Surface area comparison \|
51
52	### Interpretation
53
54	The tool provides automatic interpretation:
55	- Over-generating: Low precision, high extra geometry
56	- Under-generating: Low recall, missing geometry
57	- Size issues: Volume ratio far from 1.0
58	- Position issues: Large center offset
59
60	## CLI Options
61
62	```
63	usage: model_diff.py [-h] [-o OUTPUT_DIR] [--json] [--no-export] [--demo]
64	[reference] [generated]
65
66	positional arguments:
67	reference Reference/gold model file (STEP, BREP, or STL)
68	generated Generated/predicted model file to compare
69
70	options:
71	-o, --output-dir Output directory for GLB files (default: .)
72	--json Output only JSON metrics (for pipelines)
73	--no-export Skip exporting GLB visualization files
74	--demo Run with built-in demo models
75	```
76
77	## Output Files
78
79	When `--no-export` is not set, produces GLB files for visualization:
80
81	\| File \| Description \|
82	\|------\|-------------\|
83	\| `diff_reference.glb` \| The reference model (A) \|
84	\| `diff_generated.glb` \| The generated model (B) \|
85	\| `diff_missing.glb` \| Geometry in A but not B (under-generation) \|
86	\| `diff_extra.glb` \| Geometry in B but not A (over-generation) \|
87	\| `diff_common.glb` \| Geometry in both (correct match) \|
88
89	## Example: Training Pipeline Integration
90
91	```bash
92	# Batch evaluation
93	for gen in outputs/*.step; do
94	uvx --from build123d python model_diff.py gold.step "$gen" --json --no-export
95	done \| jq -s '{
96	avg_iou: (map(.iou) \| add / length),
97	avg_precision: (map(.precision) \| add / length),
98	avg_recall: (map(.recall) \| add / length)
99	}'
100	```
101
102	## Example: Loss Function
103
104	```python
105	# In your training code, use metrics for loss:
106	loss = (
107	(1 - metrics['iou']) * 1.0 + # Primary shape match
108	abs(1 - metrics['volume_ratio']) * 0.5 + # Scale accuracy
109	metrics['center_offset'] * 0.1 # Position accuracy
110	)
111	```
112
113	## How It Works
114
115	The tool uses boolean operations from OpenCASCADE (via build123d):
116
117	```
118	Missing = Reference - Generated (A - B)
119	Extra = Generated - Reference (B - A)
120	Common = Reference & Generated (A ∩ B)
121	Union = Reference + Generated (A ∪ B)
122
123	IoU = volume(Common) / volume(Union)
124	Dice = 2 * volume(Common) / (volume(A) + volume(B))
125	Precision = volume(Common) / volume(B)
126	Recall = volume(Common) / volume(A)
127	```
128
129	## Sample Output
130
131	```
132	=================================================================
133	3D MODEL COMPARISON REPORT
134	Reference (A) vs Generated (B)
135	=================================================================
136
137	──────────────────────────── VOLUMES ────────────────────────────
138	Reference (A): 51,433.629
139	Generated (B): 45,904.426
140	Intersection (A∩B): 42,292.031
141	Missing (A-B): 9,141.598 (17.8% of A)
142	Extra (B-A): 3,612.395 (7.9% of B)
143
144	──────────────────────── PRIMARY METRICS ───────────────

People who liked this also viewed

6 picks

Type