This commit is contained in:
M5Max128
2026-05-22 10:08:11 +08:00
8 changed files with 518 additions and 15 deletions

View File

@@ -194,6 +194,8 @@ Uses a built-in 5×7 bitmap font renderer to draw labels directly on video frame
Extract a single frame from a video as JPEG image. Uses FFmpeg `select` filter.
When `frame` is omitted, the system automatically selects the best representative frame using the TKG bridge (see algorithm below).
**Auth**: Required
**Scope**: file-level
@@ -201,7 +203,7 @@ Extract a single frame from a video as JPEG image. Uses FFmpeg `select` filter.
| Field | Type | Required | Default | Description |
|-------|------|----------|---------|-------------|
| `frame` | integer | Yes | — | Zero-based frame number to extract |
| `frame` | integer | No | auto-detect | Zero-based frame number to extract. Omit for auto-detect. |
| `x` | integer | No | — | Crop start X (left edge). Requires `y`, `w`, `h`. |
| `y` | integer | No | — | Crop start Y (top edge). Requires `x`, `w`, `h`. |
| `w` | integer | No | — | Crop width in pixels. Requires `x`, `y`, `h`. |
@@ -209,9 +211,26 @@ Extract a single frame from a video as JPEG image. Uses FFmpeg `select` filter.
All four crop params (`x`, `y`, `w`, `h`) must be provided together or omitted.
#### Example
#### Auto-detect Algorithm
When `frame` is not provided, the endpoint finds the best frame using this fallback chain:
1. **Main characters**: find the two identities with the most face detections (TMDb source)
2. **Mutual gaze**: if their face traces have a TKG `CO_OCCURS_WITH` edge with `mutual_gaze=true`, take `first_frame`
3. **Co-occurrence**: fallback to the first frame where both identities appear together
4. **Single identity**: if only one main identity exists, take its highest-quality face frame
5. **Any identity**: fallback to the best-quality face frame across all identities
6. **Error**: if no face exists, returns `404`
The selected frame is constrained to the **first half of the video** (`total_frames / 2`).
#### Examples
```bash
# Auto-detect best representative frame
curl -s "$API/api/v1/file/$FILE_UUID/thumbnail" \
-H "X-API-Key: $KEY" -o representative.jpg
# Extract frame 1000 (full frame)
curl -s "$API/api/v1/file/bd80fec92b0b6963d177a2c55bf713e2/thumbnail?frame=1000" \
-H "Authorization: Bearer $JWT" -o frame_1000.jpg
@@ -224,10 +243,104 @@ curl -s "$API/api/v1/file/bd80fec92b0b6963d177a2c55bf713e2/thumbnail?frame=1000&
#### Response
- **200**: `image/jpeg` binary data
- **404**: File not found
- **404**: File not found / No faces in file (auto-detect)
- **500**: FFmpeg error (e.g., frame number exceeds video duration)
### `GET /api/v1/file/:file_uuid/clip`
#### Technical Details
| Detail | Value |
|--------|-------|
| **Backend** | FFmpeg (`ffmpeg-full`) |
| **Filter** | `select=eq(n\,FRAME)` to select frame, optional `crop=W:H:X:Y` |
| **Output** | Single JPEG via pipe (`image2pipe`, `mjpeg` codec) |
| **Cache** | `Cache-Control: public, max-age=86400` (24h) |
| **Frame number** | Zero-based (`frame=0` = first frame of video) |
---
### `GET /api/v1/file/:file_uuid/representative-frame`
Return JSON metadata about the best representative frame for the video. Uses the same auto-detect algorithm as `GET /thumbnail` (without crop support).
**Auth**: Required
**Scope**: file-level
#### Example
```bash
curl -s "$API/api/v1/file/$FILE_UUID/representative-frame" \
-H "X-API-Key: $KEY" | jq '.'
```
#### Response (200)
```json
{
"success": true,
"file_uuid": "aeed71342a899fe4b4c57b7d41bcb692",
"frame_number": 38165,
"timestamp_secs": 1526.6,
"face_quality": 37292.97,
"main_identities": [
{
"identity_uuid": "c3545906-c82d-4b66-aa1d-150bc02decce",
"name": "Audrey Hepburn",
"face_count": 16456
},
{
"identity_uuid": "2b0ddefe-e2a9-4533-9308-b375594604d5",
"name": "Cary Grant",
"face_count": 10643
}
],
"traces": [
{
"trace_id": 919,
"identity_uuid": "2b0ddefe-e2a9-4533-9308-b375594604d5",
"name": "Cary Grant",
"x": 764,
"y": 237,
"width": 199,
"height": 199,
"confidence": 0.8426
},
{
"trace_id": 920,
"identity_uuid": "c3545906-c82d-4b66-aa1d-150bc02decce",
"name": "Audrey Hepburn",
"x": 1143,
"y": 312,
"width": 215,
"height": 215,
"confidence": 0.8068
}
]
}
```
#### Response Fields
| Field | Type | Description |
|-------|------|-------------|
| `frame_number` | integer | Selected representative frame number (primary coordinate) |
| `timestamp_secs` | float | Time in seconds (derived from `frame_number / fps`) |
| `face_quality` | float | Quality score `area × confidence` of the best face at this frame |
| `main_identities` | array | Top 2 most frequent TMDb identities in the file |
| `main_identities[].name` | string | Identity display name |
| `main_identities[].face_count` | integer | Total face detections count |
| `traces` | array | All face traces present at the selected frame |
| `traces[].trace_id` | integer | Face trace ID |
| `traces[].identity_uuid` | string or null | Matched identity UUID |
| `traces[].name` | string or null | Identity name |
| `traces[].x, y, width, height` | integer | Bounding box coordinates |
| `traces[].confidence` | float | Detection confidence (0.01.0) |
#### Error Responses
| HTTP | When |
|------|------|
| `404` | File not found / No faces in file |
| `500` | Database error |
Extract a video clip (time range) as MPEG-TS stream. Uses FFmpeg `-ss` fast seek.

View File

@@ -11,7 +11,7 @@
body { font-family: -apple-system, BlinkMacSystemFont, 'Segoe UI', Roboto, sans-serif; background: #f5f5f5; color: #333; }
#app { display: flex; min-height: 100vh; }
html, body { height: 100%; }
.sidebar { width: 260px; min-height: 100vh; background: #fff; border-right: 1px solid #ddd; padding: 20px; display: flex; flex-direction: column; }
.sidebar { width: 260px; height: 100vh; position: sticky; top: 0; overflow-y: auto; background: #fff; border-right: 1px solid #ddd; padding: 20px; display: flex; flex-direction: column; }
.sidebar h1 { font-size: 18px; margin-bottom: 16px; }
.sidebar a { display: block; padding: 6px 0; color: #0066cc; text-decoration: none; font-size: 14px; cursor: pointer; }
.sidebar a:hover { color: #003d80; }

View File

@@ -194,6 +194,8 @@ Uses a built-in 5×7 bitmap font renderer to draw labels directly on video frame
Extract a single frame from a video as JPEG image. Uses FFmpeg `select` filter.
When `frame` is omitted, the system automatically selects the best representative frame using the TKG bridge (see algorithm below).
**Auth**: Required
**Scope**: file-level
@@ -201,7 +203,7 @@ Extract a single frame from a video as JPEG image. Uses FFmpeg `select` filter.
| Field | Type | Required | Default | Description |
|-------|------|----------|---------|-------------|
| `frame` | integer | Yes | — | Zero-based frame number to extract |
| `frame` | integer | No | auto-detect | Zero-based frame number to extract. Omit for auto-detect. |
| `x` | integer | No | — | Crop start X (left edge). Requires `y`, `w`, `h`. |
| `y` | integer | No | — | Crop start Y (top edge). Requires `x`, `w`, `h`. |
| `w` | integer | No | — | Crop width in pixels. Requires `x`, `y`, `h`. |
@@ -209,9 +211,26 @@ Extract a single frame from a video as JPEG image. Uses FFmpeg `select` filter.
All four crop params (`x`, `y`, `w`, `h`) must be provided together or omitted.
#### Example
#### Auto-detect Algorithm
When `frame` is not provided, the endpoint finds the best frame using this fallback chain:
1. **Main characters**: find the two identities with the most face detections (TMDb source)
2. **Mutual gaze**: if their face traces have a TKG `CO_OCCURS_WITH` edge with `mutual_gaze=true`, take `first_frame`
3. **Co-occurrence**: fallback to the first frame where both identities appear together
4. **Single identity**: if only one main identity exists, take its highest-quality face frame
5. **Any identity**: fallback to the best-quality face frame across all identities
6. **Error**: if no face exists, returns `404`
The selected frame is constrained to the **first half of the video** (`total_frames / 2`).
#### Examples
```bash
# Auto-detect best representative frame
curl -s "$API/api/v1/file/$FILE_UUID/thumbnail" \
-H "X-API-Key: $KEY" -o representative.jpg
# Extract frame 1000 (full frame)
curl -s "$API/api/v1/file/bd80fec92b0b6963d177a2c55bf713e2/thumbnail?frame=1000" \
-H "Authorization: Bearer $JWT" -o frame_1000.jpg
@@ -224,10 +243,104 @@ curl -s "$API/api/v1/file/bd80fec92b0b6963d177a2c55bf713e2/thumbnail?frame=1000&
#### Response
- **200**: `image/jpeg` binary data
- **404**: File not found
- **404**: File not found / No faces in file (auto-detect)
- **500**: FFmpeg error (e.g., frame number exceeds video duration)
### `GET /api/v1/file/:file_uuid/clip`
#### Technical Details
| Detail | Value |
|--------|-------|
| **Backend** | FFmpeg (`ffmpeg-full`) |
| **Filter** | `select=eq(n\,FRAME)` to select frame, optional `crop=W:H:X:Y` |
| **Output** | Single JPEG via pipe (`image2pipe`, `mjpeg` codec) |
| **Cache** | `Cache-Control: public, max-age=86400` (24h) |
| **Frame number** | Zero-based (`frame=0` = first frame of video) |
---
### `GET /api/v1/file/:file_uuid/representative-frame`
Return JSON metadata about the best representative frame for the video. Uses the same auto-detect algorithm as `GET /thumbnail` (without crop support).
**Auth**: Required
**Scope**: file-level
#### Example
```bash
curl -s "$API/api/v1/file/$FILE_UUID/representative-frame" \
-H "X-API-Key: $KEY" | jq '.'
```
#### Response (200)
```json
{
"success": true,
"file_uuid": "aeed71342a899fe4b4c57b7d41bcb692",
"frame_number": 38165,
"timestamp_secs": 1526.6,
"face_quality": 37292.97,
"main_identities": [
{
"identity_uuid": "c3545906-c82d-4b66-aa1d-150bc02decce",
"name": "Audrey Hepburn",
"face_count": 16456
},
{
"identity_uuid": "2b0ddefe-e2a9-4533-9308-b375594604d5",
"name": "Cary Grant",
"face_count": 10643
}
],
"traces": [
{
"trace_id": 919,
"identity_uuid": "2b0ddefe-e2a9-4533-9308-b375594604d5",
"name": "Cary Grant",
"x": 764,
"y": 237,
"width": 199,
"height": 199,
"confidence": 0.8426
},
{
"trace_id": 920,
"identity_uuid": "c3545906-c82d-4b66-aa1d-150bc02decce",
"name": "Audrey Hepburn",
"x": 1143,
"y": 312,
"width": 215,
"height": 215,
"confidence": 0.8068
}
]
}
```
#### Response Fields
| Field | Type | Description |
|-------|------|-------------|
| `frame_number` | integer | Selected representative frame number (primary coordinate) |
| `timestamp_secs` | float | Time in seconds (derived from `frame_number / fps`) |
| `face_quality` | float | Quality score `area × confidence` of the best face at this frame |
| `main_identities` | array | Top 2 most frequent TMDb identities in the file |
| `main_identities[].name` | string | Identity display name |
| `main_identities[].face_count` | integer | Total face detections count |
| `traces` | array | All face traces present at the selected frame |
| `traces[].trace_id` | integer | Face trace ID |
| `traces[].identity_uuid` | string or null | Matched identity UUID |
| `traces[].name` | string or null | Identity name |
| `traces[].x, y, width, height` | integer | Bounding box coordinates |
| `traces[].confidence` | float | Detection confidence (0.01.0) |
#### Error Responses
| HTTP | When |
|------|------|
| `404` | File not found / No faces in file |
| `500` | Database error |
Extract a video clip (time range) as MPEG-TS stream. Uses FFmpeg `-ss` fast seek.

View File

@@ -9,7 +9,7 @@ async fn doc_redirect() -> axum::response::Redirect {
async fn wasm_doc_handler() -> Result<impl axum::response::IntoResponse, (StatusCode, &'static str)>
{
let path =
std::path::Path::new("/Users/accusys/momentry_core_0.1/docs_v1.0/doc_wasm/index.html");
std::path::Path::new("/Users/accusys/momentry_core/docs_v1.0/doc_wasm/index.html");
match tokio::fs::read_to_string(path).await {
Ok(html) => Ok(([("content-type", "text/html; charset=utf-8")], html)),
Err(_) => Err((StatusCode::NOT_FOUND, "Doc not found")),
@@ -22,7 +22,7 @@ async fn wasm_doc_file_handler(
if file.contains("..") || file.contains("//") {
return Err((StatusCode::NOT_FOUND, "Invalid path"));
}
let base = std::path::Path::new("/Users/accusys/momentry_core_0.1/docs_v1.0/doc_wasm");
let base = std::path::Path::new("/Users/accusys/momentry_core/docs_v1.0/doc_wasm");
let path = base.join(&file);
if !path.exists() || !path.starts_with(base) {
return Err((StatusCode::NOT_FOUND, "File not found"));

View File

@@ -690,7 +690,7 @@ async fn stream_video(
#[derive(Debug, serde::Deserialize)]
struct ThumbQuery {
frame: i64,
frame: Option<i64>,
x: Option<i32>,
y: Option<i32>,
w: Option<i32>,
@@ -703,6 +703,20 @@ async fn face_thumbnail(
Query(q): Query<ThumbQuery>,
) -> Result<impl IntoResponse, StatusCode> {
let videos_table = schema::table_name("videos");
let frame = match q.frame {
Some(f) => f,
None => {
let result = crate::core::processor::tkg::query_auto_representative_frame(
state.db.pool(),
&file_uuid,
)
.await
.map_err(|_| StatusCode::NOT_FOUND)?;
result.frame_number
}
};
let row: Option<(String,)> = sqlx::query_as(&format!(
"SELECT file_path FROM {} WHERE file_uuid = $1",
videos_table
@@ -713,7 +727,7 @@ async fn face_thumbnail(
.map_err(|_| StatusCode::INTERNAL_SERVER_ERROR)?;
let (file_path,) = row.ok_or(StatusCode::NOT_FOUND)?;
let select = format!("select=eq(n\\,{})", q.frame);
let select = format!("select=eq(n\\,{})", frame);
let vf = if let (Some(x), Some(y), Some(w), Some(h)) = (q.x, q.y, q.w, q.h) {
format!("{},crop={}:{}:{}:{}", select, w, h, x, y)
} else {

View File

@@ -33,6 +33,10 @@ pub fn trace_agent_routes() -> Router<crate::api::types::AppState> {
"/api/v1/file/:file_uuid/tkg/rebuild",
post(rebuild_tkg),
)
.route(
"/api/v1/file/:file_uuid/representative-frame",
get(get_representative_frame),
)
}
#[derive(Debug, Deserialize)]
@@ -783,3 +787,59 @@ async fn rebuild_tkg(
}),
}
}
// ── Representative Frame (JSON) ───────────────────────────────────
use crate::core::processor::tkg;
#[derive(Serialize)]
struct RepFrameResponse {
success: bool,
file_uuid: String,
frame_number: i64,
timestamp_secs: f64,
face_quality: f64,
main_identities: Vec<tkg::MainIdentityInfo>,
traces: Vec<tkg::FrameTraceInfo>,
}
async fn get_representative_frame(
State(state): State<crate::api::types::AppState>,
Path(file_uuid): Path<String>,
) -> Result<Json<RepFrameResponse>, (StatusCode, Json<serde_json::Value>)> {
let result = tkg::query_auto_representative_frame(
state.db.pool(),
&file_uuid,
)
.await
.map_err(|e| {
(StatusCode::NOT_FOUND, Json(serde_json::json!({"error": e.to_string()})))
})?;
let fps = query_fps(state.db.pool(), &file_uuid).await;
Ok(Json(RepFrameResponse {
success: true,
file_uuid,
frame_number: result.frame_number,
timestamp_secs: result.frame_number as f64 / fps,
face_quality: result.face_quality,
main_identities: result.main_identities,
traces: result.traces,
}))
}
async fn query_fps(pool: &sqlx::PgPool, file_uuid: &str) -> f64 {
use crate::core::db::schema;
let video_table = schema::table_name("videos");
sqlx::query_scalar(&format!(
"SELECT COALESCE(fps, 25.0) FROM {} WHERE file_uuid = $1",
video_table
))
.bind(file_uuid)
.fetch_optional(pool)
.await
.ok()
.flatten()
.unwrap_or(25.0)
}

View File

@@ -36,6 +36,9 @@ pub use scene_classification::{
SceneSegment,
};
pub use story::{process_story, StoryChildChunk, StoryParentChunk, StoryResult, StoryStats};
pub use tkg::{build_tkg, TkgResult};
pub use tkg::{
build_tkg, query_auto_representative_frame, FrameTraceInfo, MainIdentityInfo,
RepresentativeFrameResult, TkgResult,
};
pub use visual_chunk::{process_visual_chunk, process_visual_chunk_advanced, VisualChunkResult};
pub use yolo::{process_yolo, YoloFrame, YoloObject, YoloResult};

View File

@@ -1,5 +1,5 @@
use anyhow::{Context, Result};
use serde::Deserialize;
use serde::{Deserialize, Serialize};
use sqlx::PgPool;
use std::collections::HashMap;
use std::path::Path;
@@ -835,6 +835,206 @@ async fn build_face_face_edges(pool: &PgPool, file_uuid: &str, pose_data: &[Face
Ok(edge_count)
}
// ── TKG Bridge: Representative Frame ──────────────────────────────
#[derive(Debug, Serialize)]
pub struct FrameTraceInfo {
pub trace_id: i32,
pub identity_uuid: Option<String>,
pub name: Option<String>,
pub x: i32,
pub y: i32,
pub width: i32,
pub height: i32,
pub confidence: f64,
}
#[derive(Debug, Serialize)]
pub struct MainIdentityInfo {
pub identity_uuid: String,
pub name: String,
pub face_count: i64,
}
#[derive(Debug, Serialize)]
pub struct RepresentativeFrameResult {
pub frame_number: i64,
pub face_quality: f64,
pub main_identities: Vec<MainIdentityInfo>,
pub traces: Vec<FrameTraceInfo>,
}
pub async fn query_auto_representative_frame(
pool: &PgPool,
file_uuid: &str,
) -> Result<RepresentativeFrameResult> {
let id_table = t("identities");
let fd_table = t("face_detections");
let nodes_table = t("tkg_nodes");
let edges_table = t("tkg_edges");
let video_table = t("videos");
let half_frame: i64 = match sqlx::query_scalar::<_, i64>(&format!(
"SELECT COALESCE(total_frames / 2, 0) FROM {} WHERE file_uuid = $1",
video_table
))
.bind(file_uuid)
.fetch_optional(pool)
.await?
{
Some(f) if f > 0 => f,
_ => i64::MAX,
};
let mains = sqlx::query_as::<_, (i32, String, String, i64)>(&format!(
"SELECT i.id, i.uuid::text, i.name, COUNT(fd.id)::bigint \
FROM {} fd \
JOIN {} i ON i.id = fd.identity_id \
WHERE fd.file_uuid = $1 AND fd.identity_id IS NOT NULL \
AND i.source = 'tmdb' \
GROUP BY i.id, i.uuid, i.name \
ORDER BY COUNT(fd.id) DESC LIMIT 2",
fd_table, id_table
))
.bind(file_uuid)
.fetch_all(pool)
.await
.context("Failed to detect main identities")?;
let main_ids: Vec<(i32, String, String, i64)> = mains;
let main_idents: Vec<MainIdentityInfo> = main_ids.iter().map(|(_, u, n, c)|
MainIdentityInfo { identity_uuid: u.clone(), name: n.clone(), face_count: *c }
).collect();
let frame_number: Option<i64> = if main_ids.len() >= 2 {
let id_a = main_ids[0].0;
let id_b = main_ids[1].0;
let trace_a: Option<(i32,)> = sqlx::query_as(&format!(
"SELECT trace_id FROM {} WHERE file_uuid = $1 AND identity_id = $2 \
AND trace_id IS NOT NULL GROUP BY trace_id ORDER BY COUNT(*) DESC LIMIT 1",
fd_table
))
.bind(file_uuid).bind(id_a)
.fetch_optional(pool).await?;
let trace_b: Option<(i32,)> = sqlx::query_as(&format!(
"SELECT trace_id FROM {} WHERE file_uuid = $1 AND identity_id = $2 \
AND trace_id IS NOT NULL GROUP BY trace_id ORDER BY COUNT(*) DESC LIMIT 1",
fd_table
))
.bind(file_uuid).bind(id_b)
.fetch_optional(pool).await?;
match (trace_a, trace_b) {
(Some((ta,)), Some((tb,))) => {
let tkg_frame: Option<(i64,)> = sqlx::query_as(&format!(
"SELECT (e.properties->>'first_frame')::bigint \
FROM {} e \
JOIN {} a ON a.id = e.source_node_id \
JOIN {} b ON b.id = e.target_node_id \
WHERE e.file_uuid = $1 \
AND a.external_id = concat('trace_', $2) \
AND b.external_id = concat('trace_', $3) \
AND e.properties->>'mutual_gaze' = 'true' \
LIMIT 1",
edges_table, nodes_table, nodes_table
))
.bind(file_uuid).bind(ta).bind(tb)
.fetch_optional(pool).await?;
if let Some((f,)) = tkg_frame {
if f <= half_frame { Some(f) } else { None }
} else {
sqlx::query_scalar::<_, i64>(&format!(
"SELECT MIN(fd_a.frame_number)::bigint \
FROM {} fd_a \
JOIN {} fd_b ON fd_a.frame_number = fd_b.frame_number \
WHERE fd_a.file_uuid = $1 AND fd_a.identity_id = $2 \
AND fd_b.identity_id = $3 AND fd_a.frame_number <= $4",
fd_table, fd_table
))
.bind(file_uuid).bind(id_a).bind(id_b).bind(half_frame)
.fetch_optional(pool).await?
}
}
_ => None,
}
} else {
None
};
let frame_number: Option<i64> = match frame_number {
Some(f) => Some(f),
None => {
if let Some((first_id,)) = main_ids.first().map(|(id, _, _, _)| (*id,)) {
sqlx::query_scalar::<_, i64>(&format!(
"SELECT frame_number::bigint FROM {} \
WHERE file_uuid = $1 AND identity_id = $2 \
AND frame_number <= $3 \
ORDER BY (width::float8 * height::float8) * confidence::float8 DESC \
LIMIT 1",
fd_table
))
.bind(file_uuid).bind(first_id).bind(half_frame)
.fetch_optional(pool).await?
} else {
None
}
}
};
let frame_number: Option<i64> = match frame_number {
Some(f) => Some(f),
None => {
sqlx::query_scalar::<_, i64>(&format!(
"SELECT frame_number::bigint FROM {} \
WHERE file_uuid = $1 AND identity_id IS NOT NULL \
AND frame_number <= $2 \
ORDER BY (width::float8 * height::float8) * confidence::float8 DESC \
LIMIT 1",
fd_table
))
.bind(file_uuid).bind(half_frame)
.fetch_optional(pool).await?
}
};
let frame_number = frame_number.ok_or_else(|| anyhow::anyhow!("No faces found in this file"))?;
let face_quality: f64 = sqlx::query_scalar::<_, f64>(&format!(
"SELECT COALESCE(MAX((width::float8 * height::float8) * confidence::float8), 0) \
FROM {} WHERE file_uuid = $1 AND frame_number = $2",
fd_table
))
.bind(file_uuid).bind(frame_number)
.fetch_one(pool).await?;
let traces: Vec<FrameTraceInfo> = sqlx::query_as::<_, (i32, Option<String>, Option<String>, i32, i32, i32, i32, f64)>(&format!(
"SELECT fd.trace_id, i.uuid::text, i.name, fd.x, fd.y, fd.width, fd.height, fd.confidence::float8 \
FROM {} fd \
LEFT JOIN {} i ON i.id = fd.identity_id \
WHERE fd.file_uuid = $1 AND fd.frame_number = $2 AND fd.trace_id IS NOT NULL \
ORDER BY fd.trace_id",
fd_table, id_table
))
.bind(file_uuid).bind(frame_number)
.fetch_all(pool)
.await?
.into_iter()
.map(|(trace_id, identity_uuid, name, x, y, width, height, confidence)| {
FrameTraceInfo { trace_id, identity_uuid, name, x, y, width, height, confidence }
})
.collect();
Ok(RepresentativeFrameResult {
frame_number,
face_quality,
main_identities: main_idents,
traces,
})
}
// ── Tests ─────────────────────────────────────────────────────────
#[cfg(test)]