ModelTranslator: UV 减岛优化实现计划
6 任务：overlap_tol 纯函数(TDD)→_do_unwrap/main 透传→_clean_mesh 网格清理→model_decimate --max-overlap→README→端到端验证对比。 Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
This commit is contained in:
ud18010
co-authored by
Claude Opus 4.8
Claude Opus 4.8
parent
b21925be58
commit
2673cda83a
@@ -0,0 +1,382 @@
|
||||
# UV 减岛优化实现计划（--max-overlap + 减面前网格清理）
|
||||
|
||||
> **For agentic workers:** REQUIRED SUB-SKILL: Use superpowers:subagent-driven-development (recommended) or superpowers:executing-plans to implement this plan task-by-task. Steps use checkbox (`- [ ]`) syntax for tracking.
|
||||
|
||||
**Goal:** 让减面工具能通过 `--max-overlap` 放宽重叠门限选中更高 seam 角度（更少碎岛），并在减面前清理网格，从而降低 UV 岛数、回升利用率。
|
||||
|
||||
**Architecture:** 全部改动内联进 `bl_decimate.py`（Blender 脚本）与 `model_decimate.py`（CLI）。重叠容忍由模块常量升级为 keyword-default 运行时参数，经 `uv_gate_ok`/`pick_best_candidate`/`_do_unwrap` 透传；翻转门限保持严格。纯函数（门限判定）走 TDD 单测；Blender 算子集成（`_clean_mesh`、`_do_unwrap`、argv、CLI）由端到端真实 FBX 验证。所有增强单点失败降级、不中断，FBX 必产出。
|
||||
|
||||
**Tech Stack:** Python 3（标准库 + unittest）、Blender 5.0 headless（`bpy`/`bmesh`/`mathutils`）。
|
||||
|
||||
---
|
||||
|
||||
## File Structure
|
||||
|
||||
- `Tools/ModelTranslator/bl_decimate.py`（Modify）：`uv_gate_ok`/`pick_best_candidate`/`_do_unwrap` 加 `overlap_tol` 参数；新增常量 `REL_WELD` 与函数 `_clean_mesh`；`main()` 接线（argv 第 5 位 + join 后清理）。
|
||||
- `Tools/ModelTranslator/model_decimate.py`（Modify）：新增 `--max-overlap` 参数、校验、透传。
|
||||
- `Tools/ModelTranslator/tests/test_bl_decimate.py`（Modify）：`uv_gate_ok`/`pick_best_candidate` 的 `overlap_tol` 单测。
|
||||
- `Tools/ModelTranslator/README.md`（Modify）：`--max-overlap` 与网格清理说明。
|
||||
|
||||
---
|
||||
|
||||
## Task 1: `overlap_tol` 参数穿透纯函数（TDD）
|
||||
|
||||
`uv_gate_ok` 与 `pick_best_candidate` 增加 keyword-default `overlap_tol`；翻转门限不受影响。
|
||||
|
||||
**Files:**
|
||||
- Modify: `Tools/ModelTranslator/bl_decimate.py`（`uv_gate_ok` 约 :31-33、`pick_best_candidate` 约 :41-48）
|
||||
- Test: `Tools/ModelTranslator/tests/test_bl_decimate.py`
|
||||
|
||||
- [ ] **Step 1: 写失败测试**
|
||||
|
||||
在 `tests/test_bl_decimate.py` 的 `TestUvGateOk` 类内追加以下方法：
|
||||
|
||||
```python
|
||||
def test_custom_overlap_tol_allows_higher_overlap(self):
|
||||
self.assertTrue(bd.uv_gate_ok(0.0, 0.12, overlap_tol=0.15))
|
||||
self.assertFalse(bd.uv_gate_ok(0.0, 0.16, overlap_tol=0.15))
|
||||
|
||||
def test_custom_overlap_tol_does_not_relax_flip(self):
|
||||
# 放宽重叠不影响翻转判定（翻转仍按 UV_FLIP_TOL）
|
||||
self.assertFalse(bd.uv_gate_ok(0.03, 0.0, overlap_tol=0.5))
|
||||
|
||||
def test_default_overlap_tol_matches_constant(self):
|
||||
self.assertTrue(bd.uv_gate_ok(0.0, bd.UV_OVERLAP_TOL))
|
||||
self.assertFalse(bd.uv_gate_ok(0.0, bd.UV_OVERLAP_TOL + 0.01))
|
||||
```
|
||||
|
||||
在 `TestPickBestCandidate` 类内追加：
|
||||
|
||||
```python
|
||||
def test_overlap_tol_lets_more_candidates_pass(self):
|
||||
# overlap=0.12 在默认 8% 门限下不过；放宽到 0.15 后过门并被选中
|
||||
cands = [self._c(0.0, 0.12, 50, 55)]
|
||||
self.assertIsNone(bd.pick_best_candidate(cands))
|
||||
best = bd.pick_best_candidate(cands, overlap_tol=0.15)
|
||||
self.assertEqual(best["angle"], 55)
|
||||
```
|
||||
|
||||
- [ ] **Step 2: 运行测试确认失败**
|
||||
|
||||
Run:
|
||||
```bash
|
||||
cd "d:/UD/AI/AIC#Project/Tools/ModelTranslator" && python -m unittest tests.test_bl_decimate.TestUvGateOk tests.test_bl_decimate.TestPickBestCandidate -v
|
||||
```
|
||||
Expected: FAIL — `uv_gate_ok() got an unexpected keyword argument 'overlap_tol'`（及 pick_best_candidate 同理）
|
||||
|
||||
- [ ] **Step 3: 改实现**
|
||||
|
||||
将 `bl_decimate.py` 的 `uv_gate_ok` 替换为：
|
||||
|
||||
```python
|
||||
def uv_gate_ok(flipped, overlap, overlap_tol=UV_OVERLAP_TOL):
|
||||
"""UV 质量门：翻转按 UV_FLIP_TOL 固定，重叠按 overlap_tol（默认 UV_OVERLAP_TOL）。
|
||||
overlap 为 None（op 不可用）按 0 处理。"""
|
||||
return flipped <= UV_FLIP_TOL and (overlap or 0.0) <= overlap_tol
|
||||
```
|
||||
|
||||
将 `pick_best_candidate` 替换为：
|
||||
|
||||
```python
|
||||
def pick_best_candidate(candidates, overlap_tol=UV_OVERLAP_TOL):
|
||||
"""从候选 UV 指标 dict 列表选过质量门且岛数最少者；无过门候选返回 None。
|
||||
overlap_tol 覆盖重叠容忍。每个 candidate 至少含 flipped/overlap/islands。"""
|
||||
passing = [c for c in candidates
|
||||
if uv_gate_ok(c["flipped"], c["overlap"], overlap_tol)]
|
||||
if not passing:
|
||||
return None
|
||||
return min(passing, key=lambda c: c["islands"])
|
||||
```
|
||||
|
||||
- [ ] **Step 4: 运行测试确认通过**
|
||||
|
||||
Run:
|
||||
```bash
|
||||
cd "d:/UD/AI/AIC#Project/Tools/ModelTranslator" && python -m unittest tests.test_bl_decimate -v
|
||||
```
|
||||
Expected: 全部 PASS（含新增用例）
|
||||
|
||||
- [ ] **Step 5: 提交**
|
||||
|
||||
```bash
|
||||
cd "d:/UD/AI/AIC#Project" && git add Tools/ModelTranslator/bl_decimate.py Tools/ModelTranslator/tests/test_bl_decimate.py && git commit -m "ModelTranslator: uv_gate_ok/pick_best_candidate 加 overlap_tol 参数
|
||||
|
||||
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>"
|
||||
```
|
||||
|
||||
---
|
||||
|
||||
## Task 2: `_do_unwrap` 透传 overlap_tol + main() argv 解析
|
||||
|
||||
`_do_unwrap` 接收 `overlap_tol` 并用于扫描门检与择优；`main()` 解析 argv 第 5 位。
|
||||
|
||||
**Files:**
|
||||
- Modify: `Tools/ModelTranslator/bl_decimate.py`（`_do_unwrap` 与 `main()`）
|
||||
|
||||
- [ ] **Step 1: 改 `_do_unwrap`**
|
||||
|
||||
将 `bl_decimate.py` 的整个 `_do_unwrap` 替换为（仅签名、docstring、两处 gate/择优调用带上 overlap_tol，其余逻辑不变）：
|
||||
|
||||
```python
|
||||
def _do_unwrap(obj, mode, warnings, overlap_tol=UV_OVERLAP_TOL):
|
||||
"""按模式展开：seam 从 SEAM_ANGLE_DEG 起降序扫 SEAM_ANGLE_SWEEP，首个过门者即选并停扫
|
||||
——岛数随角度降单调增，故首个过门者已是过门候选里岛数最少的（pick_best_candidate 据此在
|
||||
候选集取岛数最少者，与早停一致；重跑分支为防御：早停下 best 恒为最后一档，通常不触发）；
|
||||
全失败退 smart。overlap_tol 覆盖重叠门限（翻转仍严格）。排布 UVPM4 增强，只在最终 UV
|
||||
上跑一次，失败保底内置 pack。返回 uv 指标 dict（含 mode）。"""
|
||||
if mode == "seam":
|
||||
angles = [SEAM_ANGLE_DEG] + list(SEAM_ANGLE_SWEEP)
|
||||
candidates = []
|
||||
for ang in angles:
|
||||
_unwrap_seam(obj, warnings, seam_angle_deg=ang)
|
||||
m = _collect_uv_metrics(obj)
|
||||
m["angle"] = ang
|
||||
candidates.append(m)
|
||||
if uv_gate_ok(m["flipped"], m["overlap"], overlap_tol):
|
||||
break # 该档已过门；更低角度只会更碎，无需再试
|
||||
best = pick_best_candidate(candidates, overlap_tol)
|
||||
if best is not None:
|
||||
if best["angle"] != candidates[-1]["angle"]:
|
||||
# 选中档不是最后跑的那档，重跑恢复其 UV（_unwrap_seam 会覆盖）
|
||||
_unwrap_seam(obj, warnings, seam_angle_deg=best["angle"])
|
||||
uvpm = _uvpm_repack(obj, warnings)
|
||||
m = _collect_uv_metrics(obj)
|
||||
m["mode"] = uvpm_mode_label("seam@%d" % int(best["angle"]), uvpm)
|
||||
return m
|
||||
detail = "、".join(
|
||||
"%d°(翻转%.1f%% 重叠%s)" % (
|
||||
int(c["angle"]), c["flipped"] * 100,
|
||||
"%.1f%%" % (c["overlap"] * 100) if c["overlap"] is not None else "未知")
|
||||
for c in candidates)
|
||||
warnings.append("所有 seam 角度均未过质量门（%s），回退 smart_project" % detail)
|
||||
mode = "smart_fallback"
|
||||
_unwrap_smart(obj)
|
||||
uvpm = _uvpm_repack(obj, warnings)
|
||||
m = _collect_uv_metrics(obj)
|
||||
m["mode"] = uvpm_mode_label(mode if mode == "smart_fallback" else "smart", uvpm)
|
||||
return m
|
||||
```
|
||||
|
||||
- [ ] **Step 2: 改 `main()` argv 解析与调用**
|
||||
|
||||
在 `main()` 中，`unwrap_mode = argv[3] if len(argv) > 3 else "seam"` 之后新增一行解析 overlap_tol：
|
||||
|
||||
```python
|
||||
unwrap_mode = argv[3] if len(argv) > 3 else "seam"
|
||||
overlap_tol = float(argv[4]) if len(argv) > 4 and argv[4] else UV_OVERLAP_TOL
|
||||
```
|
||||
|
||||
并把 `main()` 里对 `_do_unwrap` 的调用（现为 `uv_info = _do_unwrap(obj, unwrap_mode, warnings)`）改为：
|
||||
|
||||
```python
|
||||
uv_info = _do_unwrap(obj, unwrap_mode, warnings, overlap_tol)
|
||||
```
|
||||
|
||||
- [ ] **Step 3: 语法自检 + 单测不回归**
|
||||
|
||||
Run:
|
||||
```bash
|
||||
cd "d:/UD/AI/AIC#Project/Tools/ModelTranslator" && python -c "import ast; ast.parse(open('bl_decimate.py', encoding='utf-8').read()); print('OK')" && python -m unittest tests.test_bl_decimate 2>&1 | tail -3
|
||||
```
|
||||
Expected: `OK`，随后单测全 PASS。（`_do_unwrap` 行为在 Task 6 端到端验证。）
|
||||
|
||||
- [ ] **Step 4: 提交**
|
||||
|
||||
```bash
|
||||
cd "d:/UD/AI/AIC#Project" && git add Tools/ModelTranslator/bl_decimate.py && git commit -m "ModelTranslator: _do_unwrap 透传 overlap_tol,main 解析 argv 第5位
|
||||
|
||||
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>"
|
||||
```
|
||||
|
||||
---
|
||||
|
||||
## Task 3: 减面前网格清理 `_clean_mesh`
|
||||
|
||||
新增常量 `REL_WELD` 与 `_clean_mesh`；`main()` 在 join 后、三角化前调用。
|
||||
|
||||
**Files:**
|
||||
- Modify: `Tools/ModelTranslator/bl_decimate.py`（常量区 + `_decimate` 后新增函数 + `main()`）
|
||||
|
||||
- [ ] **Step 1: 新增常量**
|
||||
|
||||
在 `bl_decimate.py` 常量区 `STRETCH_ITERS = 30` 那一行之后追加：
|
||||
|
||||
```python
|
||||
REL_WELD = 1e-4 # 焊接距离占局部包围盒对角线比例（避免 cm/m 单位差异导致绝对阈值失准）
|
||||
```
|
||||
|
||||
- [ ] **Step 2: 新增 `_clean_mesh` 函数**
|
||||
|
||||
在 `bl_decimate.py` 的 `_decimate` 函数之后新增：
|
||||
|
||||
```python
|
||||
def _clean_mesh(obj, warnings):
|
||||
"""减面前清理：按局部包围盒对角线相对焊接重合点 + 去退化面，给 collapse 更干净的流形，
|
||||
减少被迫加的 seam。各步失败降级不中断。"""
|
||||
import bpy
|
||||
import mathutils
|
||||
bb = [mathutils.Vector(c) for c in obj.bound_box]
|
||||
diag = (bb[0] - bb[6]).length
|
||||
bpy.context.view_layer.objects.active = obj
|
||||
bpy.ops.object.mode_set(mode='EDIT')
|
||||
bpy.ops.mesh.select_all(action='SELECT')
|
||||
if diag > 0.0:
|
||||
try:
|
||||
bpy.ops.mesh.remove_doubles(threshold=diag * REL_WELD)
|
||||
except RuntimeError as e:
|
||||
warnings.append("remove_doubles 失败（%s），跳过焊接" % e)
|
||||
else:
|
||||
warnings.append("包围盒对角线为 0，跳过焊接")
|
||||
try:
|
||||
bpy.ops.mesh.dissolve_degenerate()
|
||||
except RuntimeError as e:
|
||||
warnings.append("dissolve_degenerate 失败（%s），跳过去退化" % e)
|
||||
bpy.ops.object.mode_set(mode='OBJECT')
|
||||
```
|
||||
|
||||
- [ ] **Step 3: `main()` 接线**
|
||||
|
||||
在 `main()` 中 `obj = join_meshes(meshes)` 之后、`_triangulate(obj)` 之前插入：
|
||||
|
||||
```python
|
||||
_clean_mesh(obj, warnings)
|
||||
```
|
||||
|
||||
- [ ] **Step 4: 语法自检 + 单测不回归**
|
||||
|
||||
Run:
|
||||
```bash
|
||||
cd "d:/UD/AI/AIC#Project/Tools/ModelTranslator" && python -c "import ast; ast.parse(open('bl_decimate.py', encoding='utf-8').read()); print('OK')" && python -m unittest tests.test_bl_decimate 2>&1 | tail -3
|
||||
```
|
||||
Expected: `OK`，随后单测全 PASS。（清理行为在 Task 6 端到端验证。）
|
||||
|
||||
- [ ] **Step 5: 提交**
|
||||
|
||||
```bash
|
||||
cd "d:/UD/AI/AIC#Project" && git add Tools/ModelTranslator/bl_decimate.py && git commit -m "ModelTranslator: 减面前 _clean_mesh 焊接重合点+去退化面
|
||||
|
||||
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>"
|
||||
```
|
||||
|
||||
---
|
||||
|
||||
## Task 4: `model_decimate.py` 新增 `--max-overlap`
|
||||
|
||||
CLI 参数 + 校验 + 透传给 bl_decimate（argv 第 5 位）。
|
||||
|
||||
**Files:**
|
||||
- Modify: `Tools/ModelTranslator/model_decimate.py`
|
||||
|
||||
- [ ] **Step 1: 加参数**
|
||||
|
||||
在 `model_decimate.py` 的 `ap.add_argument("--unwrap", ...)` 之后、`ap.add_argument("--blender", ...)` 之前新增：
|
||||
|
||||
```python
|
||||
ap.add_argument("--max-overlap", type=float, default=None,
|
||||
help="UV 重叠容忍上限（0-1，默认内置 0.08）；调高可让更高 seam 角度过门、减少碎岛")
|
||||
```
|
||||
|
||||
- [ ] **Step 2: 加校验**
|
||||
|
||||
在 `if args.tris <= 0: ap.error(...)` 之后新增：
|
||||
|
||||
```python
|
||||
if args.max_overlap is not None and not (0.0 < args.max_overlap <= 1.0):
|
||||
ap.error("--max-overlap 必须在 (0, 1] 内")
|
||||
```
|
||||
|
||||
- [ ] **Step 3: 透传**
|
||||
|
||||
将 `run_blender_script(blender, "bl_decimate.py", [args.input, out_fbx, str(args.tris), args.unwrap])` 改为：
|
||||
|
||||
```python
|
||||
s = run_blender_script(blender, "bl_decimate.py",
|
||||
[args.input, out_fbx, str(args.tris), args.unwrap,
|
||||
"" if args.max_overlap is None else str(args.max_overlap)])
|
||||
```
|
||||
|
||||
- [ ] **Step 4: 语法自检 + 参数校验（无需 Blender）**
|
||||
|
||||
Run:
|
||||
```bash
|
||||
cd "d:/UD/AI/AIC#Project/Tools/ModelTranslator" && python -c "import ast; ast.parse(open('model_decimate.py', encoding='utf-8').read()); print('OK')" && python model_decimate.py dummy.fbx --tris 100 --max-overlap 2 2>&1 | tail -2
|
||||
```
|
||||
Expected: `OK`；随后 argparse 报错退出，信息含 `--max-overlap 必须在 (0, 1] 内`（在启动 Blender 前就拒绝，无需真实 FBX）。
|
||||
|
||||
- [ ] **Step 5: 提交**
|
||||
|
||||
```bash
|
||||
cd "d:/UD/AI/AIC#Project" && git add Tools/ModelTranslator/model_decimate.py && git commit -m "ModelTranslator: model_decimate 新增 --max-overlap 参数（校验+透传）
|
||||
|
||||
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>"
|
||||
```
|
||||
|
||||
---
|
||||
|
||||
## Task 5: README 更新
|
||||
|
||||
**Files:**
|
||||
- Modify: `Tools/ModelTranslator/README.md`
|
||||
|
||||
- [ ] **Step 1: 更新 model_decimate.py 说明段**
|
||||
|
||||
将 README.md 的 `- **model_decimate.py**：…` 那条整体替换为（在现有基础上补入网格清理与 `--max-overlap`）：
|
||||
|
||||
```markdown
|
||||
- **model_decimate.py**：多 mesh 自动 join；**减面前清理网格**（按局部包围盒对角线相对焊接重合点 + 去退化面，给 collapse 更干净的流形、减少被迫加的 seam）；三角化后 Decimate(collapse) 减到 `--tris`（±3%，最多 2 轮修正）；旧 UV 全删后重展——默认锐边 seam 整岛展开（SLIM 展开后 `minimize_stretch` 松弛 + `average_islands_scale` 纹素均衡），质量门不达标（翻转 >2% 或重叠 >8%）先按更低 seam 角度扫描（55/45/35°）取过门且岛数最少者抢救，全失败才回退 Smart UV Project，`--unwrap smart` 可直接选投影式展开；**`--max-overlap 0.15` 放宽重叠门限**，让更高 seam 角度（岛更少）能过门被选中——岛数优先、可接受略高重叠时用（翻转门限始终严格）；排布默认用 UVPackmaster 4 重排（只在最终选定 UV 上跑一次），UVPM4 不可用自动沿用内置 pack_islands 布局；输出 `<名>_low.fbx`（`-o` 改目录）与 `<名>_low_uv.png`（UV 线框观察图，便于人工查阅切分/排布/碎岛——Blender 抽 UV 几何，系统 Python 用 Pillow 绘制，因 headless 无 GPU 无法用 Blender 直接出 PNG），日志报 UV 岛数/利用率与实际生效的 seam 角度
|
||||
```
|
||||
|
||||
- [ ] **Step 2: 提交**
|
||||
|
||||
```bash
|
||||
cd "d:/UD/AI/AIC#Project" && git add Tools/ModelTranslator/README.md && git commit -m "ModelTranslator: README 补充 --max-overlap 与减面前网格清理
|
||||
|
||||
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>"
|
||||
```
|
||||
|
||||
---
|
||||
|
||||
## Task 6: 端到端验证 + 前后对比
|
||||
|
||||
**Files:** 无（仅运行验证）
|
||||
|
||||
- [ ] **Step 1: 全量单测**
|
||||
|
||||
Run:
|
||||
```bash
|
||||
cd "d:/UD/AI/AIC#Project/Tools/ModelTranslator" && python -m unittest tests.test_bl_decimate tests.test_uv_preview tests.test_unity_assets 2>&1 | tail -3
|
||||
```
|
||||
Expected: 全部 PASS。
|
||||
|
||||
- [ ] **Step 2: 端到端 —— 放宽重叠（减岛目标）**
|
||||
|
||||
Run:
|
||||
```bash
|
||||
cd "d:/UD/AI/AIC#Project/Tools/ModelTranslator" && rm -f src/gargoyle_low_uv.png && PYTHONIOENCODING=utf-8 python model_decimate.py src/gargoyle.fbx --tris 3000 --max-overlap 0.15 2>&1 | tail -6 && ls -la src/gargoyle_low_uv.png
|
||||
```
|
||||
Expected: 成功；`模式` 为更高角度（如 `seam@55+uvpm`/`seam@45+uvpm`）；**岛数明显低于基线 686、利用率高于 38.2%**；`gargoyle_low_uv.png` 生成。记录实际 岛数/利用率/翻转/重叠/角度。用 Read 打开 PNG 确认碎岛减少、留白减少。
|
||||
|
||||
- [ ] **Step 3: 端到端 —— 默认路径不回归**
|
||||
|
||||
Run:
|
||||
```bash
|
||||
cd "d:/UD/AI/AIC#Project/Tools/ModelTranslator" && PYTHONIOENCODING=utf-8 python model_decimate.py src/gargoyle.fbx --tris 3000 2>&1 | tail -6
|
||||
```
|
||||
Expected: 成功；默认仍 8% 门限（翻转严格）；流程不崩，FBX + 观察图产出。因网格清理现默认开启，指标可能与历史基线（686/38.2%/seam@35）略有不同——记录并说明是清理带来的差异，非回归。
|
||||
|
||||
- [ ] **Step 4: 记录前后对比表**
|
||||
|
||||
在最终汇报里给出三列对比：
|
||||
- 基线（本优化前）：seam@35 / 686 岛 / 38.2% / 翻转 0% / 重叠 6.3%
|
||||
- 默认路径（清理开启，无 --max-overlap）：实测
|
||||
- --max-overlap 0.15（清理开启）：实测
|
||||
列出 模式/岛数/利用率/翻转/重叠。
|
||||
|
||||
---
|
||||
|
||||
## Self-Review 记录
|
||||
|
||||
- **Spec 覆盖**：①`--max-overlap` 穿透→Task 1(纯函数)+Task 2(_do_unwrap/main)+Task 4(CLI)；②减面前网格清理→Task 3；③测试→Task 1(TDD)+Task 6(端到端)；④README→Task 5；错误处理表（remove_doubles/dissolve_degenerate/对角线为0/越界/空串回落）→Task 2 argv 回落、Task 3 各 try/except、Task 4 校验。全部有对应任务。
|
||||
- **占位符**：无 TBD/TODO；每个代码步给出完整代码与确切命令、预期输出。
|
||||
- **类型/命名一致**：`overlap_tol`（Task 1 定义于 uv_gate_ok/pick_best_candidate，Task 2 在 _do_unwrap 使用并透传，键名一致）；`REL_WELD`/`_clean_mesh`（Task 3 定义与 main 调用一致）；`--max-overlap`→argv 第 5 位空串回落（Task 4 产生，Task 2 `main` 解析 `float(argv[4]) if ... and argv[4] else UV_OVERLAP_TOL`，契约一致）；翻转门限 `UV_FLIP_TOL` 全程不动。
|
||||
- **已知风险/取舍**：网格清理默认开启会改变默认路径指标（Task 6 Step 3 明确记录说明）；`remove_doubles` 相对阈值依赖 `obj.bound_box` 局部坐标（对角线为 0 时跳过，已处理）；`_do_unwrap` 早停使 pick_best_candidate 择优在单调假设下与早停一致（沿用既有设计，overlap_tol 只是放宽门限、不改这一性质）。
|
||||
Reference in New Issue
Block a user