[from-B] Point2RBoxV2ConsistencyLoss: loss_ssa(N,1) × keep(N,) 广播成 (N,N) 外积，square_mask 失效且数值随 N 膨胀
Nobody has claimed this yet.
Assessment
- Difficulty
- 2/5
- Estimated time
- 1-3 hours
- Newbie friendliness
- 78/100
- Issue type
- Bug
- Clarity
- Clearly specified
- Activity status
- Quiet
- Tech stack
- python, pytorch
- Domain
- machine-learning
Research direction
Start in losses/point2rbox_v2_loss.py at Point2RBoxV2ConsistencyLoss, focusing on the loss_ssa and keep/mask multiplication. Use the N=3 golden cases in tests/test_v3_head_loss.py as regression coverage. Done means the mask remains pairwise, N=3 values match the reference, and the temporary runtime workaround can be removed.
Written by the indexing model from the issue text.
Description
现象
losses/point2rbox_v2_loss.py 的 Point2RBoxV2ConsistencyLoss 中 loss_ssa 形状 (N,1) 与 keep/mask 形状 (N,) 相乘触发广播 → (N,N) 外积：
- square_cls 的 keep 掩码失效（本应逐对置零的项经外积求和后重新混入）；
- loss 数值随配对数 N 二次膨胀（实测 N=3 时 0.985 → 2.627）。
你们的 parity 没抓到是因为 rot 用例只有 1 对（N=1 时外积退化成标量，数值恰好一致）。
修复
keep.unsqueeze(-1)（或 loss_ssa squeeze 到 (N,)），一行。
参照
我在 B 仓库 tests/test_v3_head_loss.py 里有多对（N=3）的 golden 用例可直接搬来做回归；当前我测试里挂了标记 TEMP-PATCH(from-B) 的运行时 workaround（你的文件未动），你修复后我会移除。v3 的 loss_ss 分量在 N=3 下与 torch golden rel 2.8e-6（打补丁后）。
影响
v2/v3 训练中 ss 一致性 loss 的实际权重被放大且 square 类处理错误——你正在跑的 M6 v2 训练建议评估是否需要重启。
- Dominant language
- Python
- Stars
- 1
- Forks
- 0
- PR merge metrics
- No merged PRs in 30d
Contributor guide
No contributing guide indexed for this repository
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
More from VisionXLab/Point2RBox-v2-jittor
-
Difficulty 1/5 Under an hour Newbie friendliness 35/100
-
Difficulty 4/5 3-5 days Newbie friendliness 62/100
VisionXLab/Point2RBox-v2-jittor#3 · 1 comment ·
-
[from-B] networks/point2rbox_v2 sca 分支与 torchvision resized_crop 语义不符（越界截断+无 antialias，rel~1.0 量级） Open
Difficulty 3/5 1-2 days Newbie friendliness 76/100
All issues in VisionXLab/Point2RBox-v2-jittor
Similar issues
-
Difficulty 2/5 1-3 hours Newbie friendliness 88/100
-
Difficulty 2/5 1-3 hours Newbie friendliness 82/100
-
Difficulty 2/5 1-3 hours Newbie friendliness 78/100
-
enhancement
Difficulty 2/5 1-3 hours Newbie friendliness 72/100
-
Difficulty 2/5 1-3 hours Newbie friendliness 74/100