FC/.env.example at main · Allen-LPL/FC · GitHub

Repository navigation

History

# Copy to .env and fill in. .env is gitignored — never commit real credentials.

# HTTP: bind all inside the container; compose maps it to 127.0.0.1:8090 on host.

FACE_HTTP_ADDR=0.0.0.0:8090

# Optional shared secret; if set, callers must send header X-Face-Token: <value>

FACE_API_TOKEN=

# pgvector (compose service name "pgvector")

FACE_PG_DSN=postgres://face:face@pgvector:5432/face?sslmode=disable

# OSS — PUBLIC hangzhou endpoint (internal is unreachable from ys-api-001).

# Use the least-privilege face-service RAM AccessKey (NOT the main account).

OSS_ENDPOINT=https://oss-cn-hangzhou.aliyuncs.com

OSS_BUCKET=emergent-photo

OSS_ACCESS_KEY_ID=

OSS_ACCESS_KEY_SECRET=

# Aliyun facebody 1:1 fallback — only cn-shanghai is reachable.

# NOTE: the RAM user needs the SYSTEM policy AliyunVIAPIFullAccess (the custom

# facebody:CompareFace action alone returns 403 Unauthorized).

FACEBODY_REGION=cn-shanghai

FACEBODY_ENDPOINT=facebody.cn-shanghai.aliyuncs.com

# facebody only accepts standard cn-shanghai OSS domains (rejects cross-region

# OSS / CDN / custom domains). Fuzzy-band recheck copies probe+enrolled into this

# cn-shanghai staging bucket (1-day lifecycle on fbstage/) then signs shanghai

# URLs. Leave OSS_STAGING_BUCKET empty to disable staging (facebody will fail).

OSS_STAGING_ENDPOINT=https://oss-cn-shanghai.aliyuncs.com

OSS_STAGING_BUCKET=fc-fallback-sh-zj001

# facebody match cutoff (Confidence/100; facebody suggests 0.61/0.69/0.75).

FACE_FB_THRESHOLD=0.69

# SeetaFace6 models (mounted volume).

SEETA_MODEL_DIR=/models

# Recognizer model: face_recognizer_light.csta (512-dim, ~0.55 threshold) or

# face_recognizer.csta (standard, 1024-dim, higher accuracy, ~0.62 threshold).

# Switching this requires the pgvector table dropped + all faces re-extracted

# (the vector dim follows the model automatically). Production runs the 1024 model.

SEETA_RECOGNIZER=face_recognizer.csta

# Matching / liveness tuning (values below are for the 1024 standard model).

# Light model would use 0.55 / 0.45 / 0.55 instead.

FACE_MATCH_THRESHOLD=0.62

FACE_FUZZY_LOW=0.50

FACE_FUZZY_HIGH=0.62

FACE_LIVENESS_ENABLED=true

FACE_FALLBACK_ENABLED=true

FC/.env.example at main · Allen-LPL/FC · GitHub

Repository navigation

Expand file tree

Breadcrumbs

.env.example

Latest commit

History

Breadcrumbs

.env.example

File metadata and controls