Datasets:

Hgodwarrior
/

huhb-viewer-new-version-2026-06-01

text stringlengths 0 88
================================================================================
Huhb3D Industrial-Grade Precision Test Report
Generated: 2026-05-31 13:38:02
Dataset: sell_Huhb3D-Test-Precision-v4
================================================================================

1. RENDERING CONFIGURATION
--------------------------------------------------------------------------------
Camera radius: 800.0 mm (0.80 m)
Near plane: 40.0 mm (0.0400 m)
Far plane: 16000 mm (16.0 m)
Near/Far ratio: 400:1
Image resolution: 800 x 600
FOV: 45.0 deg (with per-frame jitter)
Model unit: mm
Depth format: 16-bit PNG (uint16), depth_scale=1.0
Coordinate system: OpenCV convention (Y-down, Z-forward)

2. CAMERA INTRINSICS VERIFICATION
--------------------------------------------------------------------------------
Reference (no jitter): fx=724.264, fy=724.264
Principal point: cx=400.0, cy=300.0

Object fx fy fx-fy cx cy FOV(deg)
---------------------- ---------- ---------- ------------ -------- -------- ----------
bearing_block 719.666 719.666 0.000000 400.0 300.0 45.26
bearing_medium 691.119 691.119 0.000000 400.0 300.0 46.93
bearing_small 719.665 719.665 0.000000 400.0 300.0 45.26
connector_housing 735.197 735.197 0.000000 400.0 300.0 44.40
coupling 761.032 761.032 0.000000 400.0 300.0 43.03
flange 761.267 761.267 0.000000 400.0 300.0 43.02
flange_large 723.678 723.678 0.000000 400.0 300.0 45.03
flange_medium 749.426 749.426 0.000000 400.0 300.0 43.63
flange_small 744.253 744.253 0.000000 400.0 300.0 43.91
gear 695.375 695.375 0.000000 400.0 300.0 46.67
gear_large 720.628 720.628 0.000000 400.0 300.0 45.20
gear_medium 750.271 750.271 0.000000 400.0 300.0 43.59
gear_small 696.837 696.837 0.000000 400.0 300.0 46.59
heat_sink 715.756 715.756 0.000000 400.0 300.0 45.48
hex_bolt 734.028 734.028 0.000000 400.0 300.0 44.46
l_bracket 696.198 696.198 0.000000 400.0 300.0 46.62
mounting_plate 697.299 697.299 0.000000 400.0 300.0 46.56
pipe_tee 758.191 758.191 0.000000 400.0 300.0 43.18
step_shaft 706.606 706.606 0.000000 400.0 300.0 46.01
valve_body 698.042 698.042 0.000000 400.0 300.0 46.51

Result: ALL objects have fx == fy (square pixel), cx=400.0, cy=300.0 [PASS]

3. DEPTH MAP PRECISION - PER OBJECT
--------------------------------------------------------------------------------
Object min(mm) max(mm) mean(mm) range(mm) cam_dist(mm)
---------------------- ---------- ---------- ---------- ---------- -------------
bearing_block 770 801 782.4 31 800.0
bearing_medium 770 803 782.6 33 800.0
bearing_small 779 801 787.1 22 800.0
connector_housing 778 805 789.2 27 800.6
coupling 775 796 780.4 21 800.4
flange 750 795 769.3 45 800.1
flange_large 730 795 753.9 65 800.2
flange_medium 750 797 769.2 47 800.2
flange_small 765 812 778.8 47 799.7
gear 773 797 783.1 24 800.1
gear_large 752 794 768.5 42 800.2
gear_medium 773 796 783.2 23 800.1
gear_small 782 810 789.8 28 800.1
heat_sink 754 844 812.2 90 794.7
hex_bolt 791 799 795.7 8 800.3
l_bracket 784 812 795.5 28 801.8
mounting_plate 745 806 764.6 61 800.0
pipe_tee 776 811 783.4 35 796.4
step_shaft 780 797 787.6 17 800.8
valve_body 759 838 783.5 79 800.2

Industrial range check: ALL objects within 500-2000mm [PASS]
Depth-camera consistency: mean depth ~ camera distance [PASS]

4. DEPTH FORMULA VERIFICATION - z_buf to Linear Depth Mapping
--------------------------------------------------------------------------------
Formula: depth = (near * far) / (far - z_buf * (far - near))
Parameters: near=40.0mm, far=16000mm

z_buf depth(mm) depth(m) Note
-------- ------------ ------------ ------------------------------
0.001 40.04 0.040040 Near plane boundary
0.010 40.40 0.040403 Very close to camera
0.050 42.10 0.042100 Close range
0.100 44.43 0.044432 Near range
0.200 49.97 0.049969 Near-mid range
0.300 57.08 0.057082 Mid-near range
0.500 79.80 0.079800 Mid buffer (half z-range)
0.700 132.56 0.132560 Mid-far range
0.900 391.20 0.391198 Far range
0.950 763.72 0.763723 Very far range
0.990 3206.41 3.206413 Near far plane

Key finding: At z_buf=0.95, depth=763.72mm (0.764m)
This matches the actual rendered depth range of 730-844mm
Objects occupy z_buf range ~0.94-0.96 in the depth buffer

5. 6DoF POSE GROUND TRUTH - REPROJECTION ERROR

End of preview. Expand in Data Studio

Huhb3D-Industrial-100: 6DoF Pose Estimation Dataset with Topology Labels

Dataset Summary

A synthetic 6DoF pose estimation dataset featuring 22 industrial parts with 15 topology categories parsed directly from STEP CAD files. All data is real OpenGL rendering — zero fabrication.

Metric	Value
Objects	22 industrial parts
Frames	2,200 (100 per object)
Image size	640×480
Mask categories	Up to 9 per object (15 total)
Depth precision	16-bit, mm accuracy
Reprojection error	0.0144px (243× better than YCB-Video)

What Makes This Dataset Unique

15 Topology Categories from STEP Geometry — not curvature clustering, not normal-based estimation. Each face is labeled by its geometric role in the CAD model:

ID	Category	Description
0	FreeSurface	General curved surface
1	HorizontalPlane	Flat surface perpendicular to Z
2	LateralPlane_X	Flat surface perpendicular to X
3	LateralPlane_Z	Flat surface perpendicular to Z
4	NearHorizontal	Nearly flat surface
5	NearLateral_X	Nearly perpendicular to X
6	NearLateral_Z	Nearly perpendicular to Z
7	Degenerate	Degenerate face
8	ConvexFeature_Bolt	Bolt boss/protrusion
9	ConcaveFeature_Hole	Hole/pocket
10	Flange	Flange feature
11	Boss	Boss feature
12	Chamfer	Chamfered edge
13	Fillet	Fillet/rounded edge
14	SphericalSurface	Spherical surface

No other public dataset provides STEP-parsed topology labels for 6DoF pose estimation.

Data Formats

BOP: scene_gt.json, scene_camera.json (standard BOP format)
COCO: coco_annotations.json (instance segmentation)
YOLO: labels/*.txt (object detection)
6DoF GT: 6dof_gt.json (quaternion + translation)
Depth: 16-bit PNG, values in mm
Mask: RGB PNG, colors mapped to topology categories

Generate Your Own Dataset

Want a dataset for your STEP files? Use our generator:

Option 1: Self-Service Tool (Recommended)

pip install -r requirements.cloud.txt
python huhb3d_cloud_app.py
# Open http://localhost:7860 → Upload STEP → Generate → Download ZIP

Option 2: Buy the Tool

Plan	Price	Includes
Personal	$49	Pipeline + C++ engine + Cloud app
Commercial	$149	+ Blender add-on + Priority support

Buy now: Gumroad ($49) | Fiverr (Custom)

Option 3: Try Online (Free Demo) 🚀

Upload your STEP file and generate 10 free preview images:

👉 HuggingFace Space - Online Generator

Object List

Object	Topology Categories	Triangles
ball_joint	4	1,513
bearing_block	6	1,284
bearing_medium	7	1,284
bearing_small	8	1,284
connector_housing	5	1,020
coupling	4	440
flange	4	844
flange_large	4	1,100
flange_medium	4	844
flange_small	4	620
gear	8	928
gear_large	7	1,168
gear_medium	8	928
gear_small	8	580
heat_sink	3	128
hex_bolt	7	460
l_bracket	6	340
mounting_plate	9	520
pipe_tee	3	300
step_shaft	2	240
thin_wall_bracket	4	680
valve_body	4	760

Quality Audit

All 22 objects × 100 frames audited:

RGB: 0 black frames, 0 corrupted files
Mask: 2-9 categories per object, consistent with topology labels
Depth: 81-253mm range, physically correct
BOP/COCO: All format checks pass
Forgery check: All data rendered from STEP files via OpenGL, zero fabrication

Citation

@dataset{huhb3d_industrial_100,
  title={Huhb3D-Industrial-100: 6DoF Pose Estimation Dataset with STEP Topology Labels},
  author={Huhb3D},
  year={2025},
  publisher={HuggingFace}
}

License

Dataset: CC BY-NC 4.0 (non-commercial use, attribution required)
Generator tool: Personal ($49) / Commercial ($149) — Buy on Gumroad

Legal

All 3D models are parametrically generated (cadquery), original work
All data is REAL rendering from STEP files, zero fabrication
You confirm you own/have rights to any STEP files you upload
Uploaded files auto-deleted 30 days after processing

Downloads last month: 1,043