[huntertcarver]

Camera metadata can look structurally valid while a camera-axis or pose-direction mismatch sends reconstruction rays to the wrong place. This adds an isolated, standard-library experiment that checks the camera/tap subset of measure-lab session format v2 and exports explicit OpenCV camera-to-world metadata.

It preserves tracking state and session coordinates, checks saved tap rays, and covers crop/resize intrinsics. It rejects malformed rotations, nonfinite/overflowing inputs, nonunit directions, off-image taps, duplicate records and incompatible source units. The source contract is pinned to ca89849 from #7; the experiment runs independently with synthetic fixtures and changes no production capture code.

Validation:

• python3 -m unittest -v: 14 tests passed on Python 3.13.7.
• CLI converted two synthetic frames and checked two tap rays.
• Independent review checked 100 rotations and rechecked the malformed-input fixes.
• git diff --check passed; generated outputs remain gitignored.

This validates metadata arithmetic only. It does not load photos, run a reconstruction model, validate the complete session schema, or establish physical camera accuracy. Field measurements and meter-anchor alignment remain separate work.

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

1 Skipped Deployment

Project	Deployment	Actions	Updated
house-scanning-web	Ignored	Preview	Sep 26, 2026 4:28am UTC

[coderabbitai]

Warning

Review limit reached

Next included review available in 59 minutes.

Check out review usage here.

View limit details

Limit details: You’ve used the included review currently available.

You've used all free OSS reviews for now. Wait for the free limit to reset to keep reviewing this public repository.

Learn how review limits work.

Review configuration:

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Advanced

Run ID: 3db0f8fe-500d-40b7-a479-55fded47ad36

📥 Commits

Reviewing files that changed from the base of the PR and between eb9d480 and 4b6c430.

📒 Files selected for processing (4)

• experiments/capture-contract/README.md
• experiments/capture-contract/capture_contract.py
• experiments/capture-contract/fixtures/synthetic-session.json
• experiments/capture-contract/test_contract.py

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[SamGu-NRX]

Reviewed the camera-axis conversion, saved-frame rays, crop/resize intrinsics, and the session-v2 contract against #7 at ca89849. I found no actionable correctness defect in this head. The experiment does not establish device accuracy or reconstruction interoperability, as its PR description notes. This is a draft, so this review is not a readiness recommendation.

[SamGu-NRX]

@greptileai

[greptile-apps]

Confidence Score: 5/5

[Low risk] Adds isolated experiment for camera metadata validation.

The PR appears safe to merge, though the tap image-boundary check and three previously reported quality issues remain.

Findings

1.  Off-image taps pass validation ▶
2.  Tap IDs can collide ▶
3.  Resized intrinsics can overflow ▶
4.  Output writes can destroy results ▶

Fix with agent prompt

### Issue 1
experiments/capture-contract/capture_contract.py:93
A tap at `u == width` or `v == height` passes this check, even though those coordinates lie beyond the last pixel of the saved image. If its saved ray matches the computed ray, the probe reports the off-image tap as valid, making the tap check misleading. Use an exclusive upper bound.

```suggestion
    if not (0 <= u < width and 0 <= v < height):
```

### Issue 2
experiments/capture-contract/capture_contract.py:undefined-151
Two taps with the same ID both pass validation and produce `tap_checks` entries that cannot be distinguished by ID. Missing IDs have the same problem. Reject duplicate and missing tap IDs so each reported check can be matched to its source tap.

### Issue 3
experiments/capture-contract/capture_contract.py:undefined-117
The inputs are checked for finiteness, but the computed values are not. For example, `fx=1e200` and `sx=1e200` return an infinite focal length, so a caller can receive an invalid camera matrix without an error. Check the computed values before returning them.

```suggestion
    return list(vector([fx*sx, fy*sy, (cx-x0)*sx, (cy-y0)*sy],
                       4, "resized intrinsics"))
```

### Issue 4
experiments/capture-contract/capture_contract.py:undefined-208
Writing directly to `--out` truncates an existing conversion before the new one is complete. If the write is interrupted or the disk fills, the previous result is lost and a partial JSON file remains. Write a temporary file and replace the destination after the write succeeds.

---

For each issue above, determine whether it is valid and should be fixed. If so, fix it directly.

Summary

The PR adds an isolated camera-metadata probe that converts measure-lab v2 poses to OpenCV camera-to-world matrices and checks saved tap rays.

• Includes synthetic fixtures and tests for coordinate conversion, intrinsics, and malformed inputs.
• Does not change production capture or establish physical camera accuracy.

Diagram

%%{init: {'theme': 'neutral'}}%%
flowchart LR
  A[Session JSON] --> B[Validate frames and taps]
  B --> C[Convert camera basis]
  C --> D[Output metadata and tap checks]

Loading

_{Reviews (2) · Last reviewed commit: "experiments: validate camera metadata fo..."}

[chatgpt-codex-connector]

Codex usage limits have been reached for code reviews. Please check with the admins of this repo to increase the limits by adding credits.
Repo admins can enable using credits for code reviews in their settings.

[codemetrics-ai]

Note

Review skipped

Codemetrics is installed on this repository, but it hasn't been added to your team yet.

To enable AI code reviews for this repo, go to your dashboard and add house-scanning-master to your team's repositories.