Repository navigation
[NM-571] 학습 라벨 전송 — imageId(v7) · box/score · 선택마다 라벨 · 지운 번호 - #87
Merged
Merged
Conversation
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
… 완료 때 지운 번호 - 사진마다 UUID v7 을 만들어 upload-url 본문과 속성 추출에 같은 값을 보낸다 - 속성 추출에 원본 크기 · segModel, 알약마다 box(원본 픽셀) · score 를 싣는다 - 「선택」하고 돌아올 때 PUT labels — model 은 그 알약의 첫 조회, userInput 은 마지막 조회. 수동 추가는 100 부터 · manual - 「완료」 · 「종료」 때 PATCH segmentation 으로 지운 모델 알약 번호를 보낸다 - pillId 를 0부터로 — 라벨 번호(pillIndex 0~99)와 같아야 한다 - 모두 베스트 에포트: 실패를 화면에 알리지 않고 다시 보내지 않는다 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- model: 처음 나간 요청 대신 사용자 수정을 뺀 상태로 다시 조립한다(각인이 늦게 오거나 입력 대기 중 수정에 흔들리지 않게) - userInput: 응답이 와 화면 목록이 된 요청 — 토큰 없이 다시 보냈으면 다시 보낸 쪽(CandidateAnswer) - 수동 추가 키 카운터를 세션 상태로 — 사진마다 100 부터 다시 센다 - 학습용 원본은 사진마다 한 번만 올린다(속성 추출 다시 시도에 반복하지 않게) - uploadUrl · RawImageUploader KDoc 을 키로 이어지는 설계로 고친다 - 지운 번호 계산의 쓸모없는 거르기를 뺀다 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
계약상 PUT labels · PATCH segmentation 도 AppCheckToken 대상인데 고정 경로 목록에 없어 401 APP_CHECK_FAILED 로 조용히 버려졌다(실기기 dev 서버). 경로에 변수가 있는 대상을 정규식으로 맞춘다. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Contributor
Author
통과 기준 대조
|
Merged
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
식별할 때 학습용 라벨을 서버로 보낸다. 계약은 spec
api/domains/label.md·openapi.yaml0.31.0(NM-568). 백엔드 NM-569 는 main 배포 완료.바꾼 것
imageId= UUID v7 (앱이 만듦)POST /api/v0/pill-images/upload-url본문{ imageId }— 사진마다 한 번imageId·imageWidth·imageHeight·segModel, 알약마다box(원본 픽셀) ·score— 찾은 알약 전부PUT /api/v0/pill-images/{imageId}/labels/{n}—pillCode·model·userInput. 수동 추가는n100부터 ·manual: truePATCH /api/v0/pill-images/{imageId}/segmentation— 지운 모델 알약 번호모두 베스트 에포트 — 실패를 화면에 알리지 않고 다시 보내지 않는다(
PillLabelSender).먼저 확인한 것 (티켓 「먼저 확인」)
DetectedPill.box로 내보내고, 대조 테스트에 「box 크기 = 크롭 크기」를 더했다.pillId가 1부터였다 — 검출 100개째가"100"으로 범위(0~99)를 벗어난다. 0부터로 바꿨다(내부 키 — 화면 번호는 그대로).계약 해석 — 리뷰가 필요한 자리
model: 계약 문구는 「첫 후보 조회 요청 그대로(사용자가 고치기 전)」. 실제로 처음 나간 요청을 붙잡으면 ① 온디바이스 각인 · 마크가 0.9초쯤 늦게 와서 첫 요청에 빠지고 ② 입력 대기(250ms) 중 수정하면 첫 요청이 이미 고친 값이다. 그래서 「선택」 시점에 사용자 수정을 뺀 상태로 같은 조립을 다시 해 만든다 — 모델 값 · 읽은 각인 · 마크 그대로. 고친 게 없으면userInput과 같다.userInput: 응답이 와 화면 목록이 된 요청 — 토큰이 상해 토큰 없이 다시 보냈으면 다시 보낸 쪽.segModel: 계약에 예시가 없어 마크(20260925-convnext-species)처럼 ML 폴더 이름20260610-rfdetr-seg-small로 정했다. iOS 도 같은 값이어야 한다.실기기에서 잡은 결함
라벨 두 경로가 App Check 대상(계약
AppCheckToken)인데 앱의 고정 경로 목록에 없어 401APP_CHECK_FAILED로 조용히 버려졌다. 변수 경로를 정규식으로 맞췄다(494f64c).확인
imageId본문 50B)labels/0labels/100segmentation[2]5fb22ec)🤖 Generated with Claude Code