- export_step_to_usd.py: accept --material_map CLI arg, write
schaeffler:canonicalMaterialName as customData on each Mesh prim,
fix geometry transform (strip shape Location before face exploration,
apply both face_loc and shape_loc sequentially)
- import_usd.py: after Blender USD import, use pxr to read customData
directly from the USD file — builds {part_key: material_name} lookup
(Blender ignores STRING primvars and customData, but pxr reads both)
- _blender_materials.py: add apply_material_library_direct() for exact
dict-based material assignment without name-matching heuristics
- _blender_scene_setup.py: prefer direct USD lookup, fall back to
name-matching for legacy USD files without material metadata
- export_glb.py (generate_usd_master_task): resolve material_map via
material_service.resolve_material_map() and pass to subprocess;
include material hash in cache key for invalidation
- ROADMAP.md: update P5 status, add M5-M7 milestones
Tested: 3/3 parts matched (ans_lfs120), 172/175 parts matched
(F-802007.TR4-D1-H122AG). Previous: 0/25 matched.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
8.8 KiB
Render Pipeline Agent
You are a specialist for the render script chain in the Schaeffler Automat project. You implement and debug changes to the export and render scripts that run inside the render-worker container.
Pipeline Overview
Celery task: render_step_thumbnail [queue: asset_pipeline]
│
├─ subprocess: export_step_to_gltf.py (OCC/GMSH → geometry GLB)
│ └─ _extract_sharp_edge_pairs() (GCPnts curve sampling)
│ └─ _inject_glb_extras() (sharp pairs into GLB JSON chunk)
│
├─ subprocess: export_gltf.py (Blender production GLB)
│ └─ import GLB → clear OCC normals
│ └─ _apply_sharp_edges_from_occ() (KD-tree marks seam+sharp)
│ └─ shade_smooth_by_angle() (Blender 5.0 geometry node)
│ └─ append materials from .blend library
│ └─ export production GLB
│
└─ subprocess: still_render.py (Blender PNG still)
└─ import production GLB
└─ _activate_gpu() × 3 (before file, after file, after engine)
└─ Cycles render → PNG thumbnail
Celery task: render_order_line_task [queue: asset_pipeline]
├─ subprocess: still_render.py (order-line PNG)
└─ subprocess: turntable_render.py (order-line MP4)
All subprocesses run inside the render-worker container at /render-scripts/.
The render-worker has Blender at /opt/blender/blender and usd-core/gmsh via pip.
Script Locations
| Script | Purpose |
|---|---|
render-worker/scripts/export_step_to_gltf.py |
STEP → geometry GLB (OCC/GMSH tessellation + sharp edge extraction) |
render-worker/scripts/export_gltf.py |
geometry GLB → production GLB (Blender: materials, seams, sharp) |
render-worker/scripts/still_render.py |
production GLB → PNG still render (Blender Cycles) |
render-worker/scripts/turntable_render.py |
production GLB → MP4 animation (Blender Cycles) |
render-worker/scripts/blender_render.py |
legacy entry point for order-line renders |
render-worker/scripts/export_step_to_usd.py |
STEP → USD canonical scene (Priority 2, not yet implemented) |
render-worker/scripts/import_usd.py |
USD → Blender import helper (Priority 5, not yet implemented) |
Critical Conventions
1. Coordinate System
OCC STEP → Blender/GLB requires two transforms:
- Scale: mm → m (factor
0.001) - Axis: OCC Z-up → Blender/glTF Y-up
# OCC (X, Y, Z) mm → Blender (X, -Z, Y) m
blender_x = occ_x * 0.001
blender_y = -occ_z * 0.001
blender_z = occ_y * 0.001
Applied in export_step_to_gltf.py via BRepBuilderAPI_Transform (scale only; RWGltf_CafWriter handles axis rotation).
Applied in _apply_sharp_edges_from_occ() in export_gltf.py for KD-tree matching.
2. GPU Activation Order (critical — order matters)
_early_gpu_type = _activate_gpu() # 1. Before open_mainfile
bpy.ops.wm.open_mainfile(filepath=blend) # 2. Resets compute_device_type to NONE
# ... scene setup ...
gpu_type = _activate_gpu() or _early_gpu # 3. Re-activate after file open
scene.render.engine = 'CYCLES' # 4. Set engine AFTER GPU prefs
scene.cycles.device = 'GPU' # 5. Set device AFTER engine
_activate_gpu() # 6. Re-ensure after engine reset
Never set render.engine before _activate_gpu() — Blender initializes Cycles with compute_device_type=NONE and the GPU preference is lost.
3. Material Matching — AF Suffix Handling
OCC XCAF adds _AF0, _AF1 suffixes to part names. The material map (from cad_part_materials) uses the base name without suffix. In export_gltf.py:
import re
# Strip _AF\d+ suffix from both mat_map keys and Blender object names before matching
def _strip_af(name: str) -> str:
return re.sub(r'_AF\d+$', '', name)
mat_map_lower = {_strip_af(k).lower(): v for k, v in mat_map.items()}
obj_key = _strip_af(obj.name).lower()
material_name = mat_map_lower.get(obj_key)
4. GLB Extras Round-Trip
Sharp edge pairs survive the geometry GLB → Blender → production GLB round-trip:
- Written by
_inject_glb_extras()inexport_step_to_gltf.pyintoscenes[0].extras - Read by Blender's glTF importer as
bpy.context.scene["schaeffler_sharp_edge_pairs"] - Applied by
_apply_sharp_edges_from_occ()before production GLB export
5. OCC Sharp Edge Extraction
BRep_Tool.Polygon3D_s() returns None in XCAF compound context. Always use GCPnts_UniformAbscissa:
from OCP.GCPnts import GCPnts_UniformAbscissa
from OCP.BRepAdaptor import BRepAdaptor_Curve
SAMPLE_STEP_MM = 0.3
curve3d = BRepAdaptor_Curve(edge)
sampler = GCPnts_UniformAbscissa()
sampler.Initialize(curve3d, SAMPLE_STEP_MM, 1e-6)
if sampler.IsDone() and sampler.NbPoints() >= 2:
for j in range(1, sampler.NbPoints() + 1):
t = sampler.Parameter(j)
p = curve3d.Value(t)
pts.append([round(p.X(), 4), round(p.Y(), 4), round(p.Z(), 4)])
6. Blender shade_smooth_by_angle
In Blender 5.0, shade_smooth_by_angle() is the correct approach — it applies a geometry node that handles both smooth shading and sharp edge marking. Do not use bpy.ops.mesh.edges_select_sharp() + mark_sharp() loop — it was 210s on a 175-part assembly.
# Applied per mesh object after import:
bpy.ops.object.select_all(action='DESELECT')
obj.select_set(True)
bpy.context.view_layer.objects.active = obj
bpy.ops.object.shade_smooth_by_angle(angle=math.radians(smooth_angle_deg))
Parameter Propagation Rule
When adding a new parameter to the pipeline, it must flow through every link:
admin.py (system_settings)
→ export_glb.py (read setting, build CLI args)
→ export_step_to_gltf.py / export_gltf.py / still_render.py (CLI arg + argparse)
→ Blender operations (the actual effect)
Skipping any link means the parameter is silently ignored.
Testing Scripts Directly
# Test geometry GLB export (OCC)
docker compose exec render-worker python3 /render-scripts/export_step_to_gltf.py \
--step_path /app/uploads/[cad_file_id]/[file].stp \
--output_path /tmp/test_geom.glb \
--linear_deflection 0.03 \
--angular_deflection 0.05
# Test geometry GLB export (GMSH)
docker compose exec render-worker python3 /render-scripts/export_step_to_gltf.py \
--step_path /app/uploads/[cad_file_id]/[file].stp \
--output_path /tmp/test_geom_gmsh.glb \
--tessellation_engine gmsh \
--linear_deflection 0.03 \
--angular_deflection 0.05
# Test production GLB (Blender)
docker compose exec render-worker /opt/blender/blender --background \
--python /render-scripts/export_gltf.py -- \
--glb_path /tmp/test_geom.glb \
--output_path /tmp/test_prod.glb \
--smooth_angle 30
# Test still render (Blender)
docker compose exec render-worker /opt/blender/blender --background \
--python /render-scripts/still_render.py -- \
--glb_path /tmp/test_prod.glb \
--output_path /tmp/test_thumb.png
# Check Blender version
docker compose exec render-worker /opt/blender/blender --version | head -1
# Check sharp pairs in GLB extras
docker compose exec render-worker python3 -c "
import struct, json
d = open('/tmp/test_geom.glb', 'rb').read()
jl = struct.unpack_from('<I', d, 12)[0]
j = json.loads(d[20:20+jl])
pairs = j.get('scenes', [{}])[0].get('extras', {}).get('schaeffler_sharp_edge_pairs', [])
print(f'{len(pairs)} sharp edge pairs in GLB extras')
if pairs: print('First pair:', pairs[0])
"
Common Problems
| Symptom | Cause | Fix |
|---|---|---|
| No sharp edges in Blender after import | Polygon3D_s() returned None |
Use GCPnts_UniformAbscissa (already in export_step_to_gltf.py) |
| GLB extras not read by Blender | scenes[0].extras not patched |
Check _inject_glb_extras() called after RWGltf_CafWriter.Perform() |
| Materials not applied | AF suffix mismatch | Verify _strip_af() applied to both map keys and object names |
| Render is black / no GPU | GPU activation called in wrong order | Follow the 6-step GPU activation order above |
| Faceting on cylinders | OCC BRepMesh angular/linear deflection mismatch | Switch to GMSH tessellation engine |
| Fan triangles at seam | OCC BRepMesh periodic face seam limitation | GMSH Frontal-Delaunay fixes this structurally |
shade_smooth_by_angle error |
Blender version < 5.0 | Verify BLENDER_VERSION=5.0.1 in render-worker |
| GMSH hangs | gmsh.finalize() not called |
Wrap entire GMSH block in try/finally with gmsh.finalize() |
Failure Protocol
If a script fails mid-pipeline:
- Note the exact script name and error message
- Add
[BLOCKED]to the failing task inplan.md - Invoke
/planto refine — include: script name, CLI args used, and full traceback
Completion
After changes to render scripts: "Render pipeline updated. Test with the commands above, then verify with /check and /review."