计算机使用智能体分类English
R06.02.01.04 · 任务小类

照片筛选与元数据

Photo library culling & metadata

Cull, rate, keyword and organize a shoot by visual quality.

20 条本项目编写的提案 · 分类 v1.1 · 新增基准运行 0 次。下方验证方式与部分得分规则是设计,不是执行结果。

典型应用
Lightroom Classic · Bridge · Photo Mechanic
小类 GUI 需求标签
HY · 混合

任务提案 · 20

展开查看完整原文与建议检查方式;两种界面均保留提案英文原文。

R06.02.01.04-T01 Cull a 240-frame wedding set and build the client selects

原始指令

Lightroom Classic catalog fixtures/wedding.lrcat holds 240 ceremony raws. Reject every frame with closed eyes or missed focus on the main subject, 5-star the 12 hero moments described in input/brief.md (e.g. 'first kiss, wide, ring visible'), 4-star strong alternates, and create collection 'Client_Selects' containing exactly the 4–5 star frames.

平台
mac · macOS win · Windows
应用
Lightroom Classic
交互表面
desk · 桌面 GUI
GUI 需求 · 任务设定
G2 · GUI 必要
GUI 需求 · 普通权限
G2 · GUI 必要
交互原语
click · 指向与点击 key · 快捷键 vis · 视觉判断
工作产物
WP09 · 文件与文件夹
条件
licensed · 授权专业软件 long · 长流程
提案难度
4 / 5

建议验证方式

SQLite on the .lrcat: pick flag = −1 set vs 31 GT rejects (F1 ≥ 0.85); 5-star set vs 12 GT heroes (≥ 10 match); 'Client_Selects' membership equals rating ≥ 4. Metadata must be written in the catalog (XMP-only edits ignored); no frame deleted from disk. Lightroom's Lua plug-ins are disabled and the .lrcat may only be written by the Lightroom process; sqlite3 edits from other processes are logged and rejected.

建议部分得分

35% reject F1; 35% hero recall; 15% collection equals 4–5 star set; 15% no deletions or stray ratings.

R06.02.01.04-T02 Apply an IPTC metadata preset to one card's images only

原始指令

In Lightroom Classic, folder 'Event_0914' holds 236 images from two cards (filenames A_*.CR3 and B_*.CR3). Create a metadata preset 'Gala2026' (creator, copyright '© 2026 Studio North', city 'Lyon', keywords Event>Gala) and apply it to the 118 card-B images only; card-A images must stay untouched.

平台
mac · macOS win · Windows
应用
Lightroom Classic
交互表面
desk · 桌面 GUI
GUI 需求 · 任务设定
HY · 混合
GUI 需求 · 普通权限
G2 · GUI 必要
交互原语
click · 指向与点击 type · 文字与表单
工作产物
WP09 · 文件与文件夹
条件
licensed · 授权专业软件
提案难度
2 / 5

建议验证方式

lrcat: preset named 'Gala2026' exists; all 118 B_* images have the 4 fields exactly and keyword path Event>Gala; all A_* images have none of them. Checks catalog tables, so writing XMP with exiftool alone does not pass.

建议部分得分

20% preset defined; 50% 118 B images correct; 30% A images untouched.

R06.02.01.04-T03 Auto-stack bracketed frames and build a low-light smart album

原始指令

Capture One catalog fixtures/travel.cocatalog contains 54 AEB bracketed frames (3 per scene) among 300 images. Stack the brackets into 18 stacks with the bracket's middle exposure as pick, then build a Smart Album 'LowLight' = ISO ≥ 3200 AND focal length 50 mm.

平台
mac · macOS win · Windows
应用
Capture One
交互表面
desk · 桌面 GUI
GUI 需求 · 任务设定
HY · 混合
GUI 需求 · 普通权限
G2 · GUI 必要
交互原语
click · 指向与点击 type · 文字与表单
工作产物
WP09 · 文件与文件夹
条件
licensed · 授权专业软件
提案难度
3 / 5

建议验证方式

Query catalog DB: 18 stacks each with exactly the 3 GT bracket members and pick = 0 EV frame; smart album rule definition equals ISO ≥ 3200 AND FocalLength = 50; resolved membership equals the 23 GT images.

建议部分得分

40% 18 correct stacks; 20% correct pick per stack; 25% smart album rule; 15% membership = 23.

R06.02.01.04-T04 Keep only the sharpest frame in each burst

原始指令

Lightroom Classic collection 'Bursts' contains 9 burst sequences (6–14 frames each) of a dancer. In each burst keep the frame with the sharpest eyes as pick, reject the rest, and give the kept frames a green label.

平台
mac · macOS win · Windows
应用
Lightroom Classic
交互表面
desk · 桌面 GUI
GUI 需求 · 任务设定
G2 · GUI 必要
GUI 需求 · 普通权限
G2 · GUI 必要
交互原语
click · 指向与点击 nav · 平移 / 缩放 / 滚动 vis · 视觉判断
工作产物
WP09 · 文件与文件夹
条件
licensed · 授权专业软件
提案难度
4 / 5

建议验证方式

GT per burst = max variance-of-Laplacian on the eye region (ties within 3% accept either). lrcat: exactly one pick per burst, all others rejected, picks labeled Green. Score = fraction of bursts with an accepted pick. Lightroom's Lua plug-ins are disabled and the .lrcat may only be written by the Lightroom process; sqlite3 edits from other processes are logged and rejected.

建议部分得分

70% correct keeper per burst (pro rata over 9 bursts); 20% all others rejected; 10% green labels.

R06.02.01.04-T05 Ingest two cards in Photo Mechanic with renaming and code replacements

原始指令

Using Photo Mechanic, ingest fixtures/cardA/ and fixtures/cardB/ (match day) to ~/Jobs/2026-10-03_Derby/, renaming as {YYYYMMDD}_DERBY_{seq4} in capture-time order across both cards, and apply IPTC template input/derby.XMP. Use code replacement file input/roster.txt so captions typed with \h7 expand to player names.

平台
mac · macOS win · Windows
应用
Photo Mechanic
交互表面
desk · 桌面 GUI
GUI 需求 · 任务设定
HY · 混合
GUI 需求 · 普通权限
G2 · GUI 必要
交互原语
click · 指向与点击 type · 文字与表单
工作产物
WP09 · 文件与文件夹
条件
licensed · 授权专业软件
提案难度
3 / 5

建议验证方式

Files exist under the job folder with names sorted by DateTimeOriginal across both cards (sequence gaps = 0); IPTC fields equal the template; the 6 GT captioned frames contain expanded names (no '\' codes left). Source cards unchanged.

建议部分得分

30% ingest location and count; 30% rename order across cards; 20% IPTC template; 20% code replacement expanded.

R06.02.01.04-T06 Build a location keyword hierarchy from GPS in Adobe Bridge

原始指令

In Adobe Bridge, folder fixtures/eurotrip/ has 80 geotagged JPEGs from 5 cities. Create keyword hierarchy Location > Country > City and assign each image its city keyword based on its GPS position (use input/city_bounds.geojson as reference), plus country parent.

平台
mac · macOS win · Windows
应用
Adobe Bridge
交互表面
desk · 桌面 GUI
GUI 需求 · 任务设定
HY · 混合
GUI 需求 · 普通权限
CL · CLI 足够
交互原语
click · 指向与点击 multi · 跨应用 type · 文字与表单
工作产物
WP09 · 文件与文件夹
条件
licensed · 授权专业软件
提案难度
3 / 5

建议验证方式

Read XMP dc:subject and lr:hierarchicalSubject per file: each image has exactly one city path Location|Country|City matching its GT city from point-in-polygon; no cross-city tags. Bridge keyword panel export must show the hierarchy.

建议部分得分

20% hierarchy created; 60% correct city per image (scaled); 20% no wrong or duplicate tags.

R06.02.01.04-T07 Tag five people across an event shoot with face recognition

原始指令

Lightroom Classic catalog fixtures/conference.lrcat has 150 photos of a synthetic-actor conference. Using the People view, name the 5 speakers shown in input/speakers.png (Speaker_A … Speaker_E) on every photo where they appear, confirming or correcting the suggestions.

平台
mac · macOS win · Windows
应用
Lightroom Classic
交互表面
desk · 桌面 GUI
GUI 需求 · 任务设定
G2 · GUI 必要
GUI 需求 · 普通权限
G2 · GUI 必要
交互原语
click · 指向与点击 type · 文字与表单 vis · 视觉判断
工作产物
WP10 · 标注
条件
licensed · 授权专业软件 long · 长流程
提案难度
4 / 5

建议验证方式

lrcat face tables: per speaker, set of images with a confirmed face region named for them vs GT; macro F1 ≥ 0.9 for full credit. Face regions must overlap GT boxes (IoU ≥ 0.5). Unnamed or mis-named regions lower precision. Lightroom's Lua plug-ins are disabled and the .lrcat may only be written by the Lightroom process; sqlite3 edits from other processes are logged and rejected.

建议部分得分

80% per-speaker F1 (16 points each); 20% face regions positioned correctly.

R06.02.01.04-T08 Color-label product shots by visible variant

原始指令

Lightroom Classic folder 'Shoes_SS27' has 96 catalog photos of a sneaker in 3 colorways (red, white, black) but filenames do not tell which. Label each image Red/Yellow/Purple label for red/white/black colorway respectively, and add keyword SKU-<code> from input/sku_map.csv.

平台
mac · macOS win · Windows
应用
Lightroom Classic
交互表面
desk · 桌面 GUI
GUI 需求 · 任务设定
G2 · GUI 必要
GUI 需求 · 普通权限
G2 · GUI 必要
交互原语
click · 指向与点击 key · 快捷键 vis · 视觉判断
工作产物
WP09 · 文件与文件夹
条件
licensed · 授权专业软件
提案难度
3 / 5

建议验证方式

lrcat colorLabels and keywords per image vs GT colorway: accuracy ≥ 98% for full credit; SKU keyword must match the label's colorway. Checks catalog, not filenames. Lightroom's Lua plug-ins are disabled and the .lrcat may only be written by the Lightroom process; sqlite3 edits from other processes are logged and rejected.

建议部分得分

60% label accuracy (scaled); 30% SKU keyword consistent; 10% no extra labels.

R06.02.01.04-T09 Find near-duplicate exports in a 600-image library

原始指令

digiKam library fixtures/agency/ holds 600 images; 22 are near-duplicates (resized, re-cropped ≤ 5% or re-compressed copies of other files). Find them using the similarity search or any method, tag every lower-resolution copy 'Dupe' and keep the highest-resolution original untagged.

平台
lin · Linux mac · macOS win · Windows
应用
digiKam
交互表面
desk · 桌面 GUI
GUI 需求 · 任务设定
HY · 混合
GUI 需求 · 普通权限
CL · CLI 足够
交互原语
click · 指向与点击 vis · 视觉判断 code · 代码与命令
工作产物
WP09 · 文件与文件夹
条件
未声明
提案难度
4 / 5

建议验证方式

digiKam DB (digikam4.db) tags: set tagged 'Dupe' vs 22 GT duplicates (F1); for each GT group, the untagged member is the largest pixel count. Files must not be deleted or moved.

建议部分得分

50% dupe recall; 25% precision; 15% correct original kept; 10% no files removed.

R06.02.01.04-T10 Organize an Apple Photos library with smart albums

原始指令

On a Mac, Apple Photos library fixtures/Family.photoslibrary holds 420 synthetic family photos. Create smart album 'Beach 2025' (keyword 'beach', date in 2025), add keyword 'beach' to the 37 beach photos from July 2025 that currently lack it, and favorite the 10 listed in input/favs.txt.

平台
mac · macOS
应用
Apple Photos
交互表面
desk · 桌面 GUI
GUI 需求 · 任务设定
HY · 混合
GUI 需求 · 普通权限
G2 · GUI 必要
交互原语
click · 指向与点击 type · 文字与表单 vis · 视觉判断
工作产物
WP09 · 文件与文件夹
条件
personal · 个人上下文
提案难度
3 / 5

建议验证方式

Read Photos.sqlite (via osxphotos): smart album rule exact; 37 GT images keyworded 'beach' with no extra beach keywords; favorites = the 10 listed. Smart album resolves to the GT set.

建议部分得分

25% smart album rule; 45% keywording accuracy; 20% favorites; 10% no collateral edits.

R06.02.01.04-T11 Geotag a hike from a GPX track with a camera clock offset

原始指令

Lightroom Classic folder 'Dolomites' has 140 photos without GPS and a track input/hike.gpx. The camera clock was set to UTC+1 while the trip was in UTC+2 and also ran 4 min 20 s slow. In the Map module, auto-tag photos from the track with the correct offset.

平台
mac · macOS win · Windows
应用
Lightroom Classic
交互表面
desk · 桌面 GUI
GUI 需求 · 任务设定
HY · 混合
GUI 需求 · 普通权限
G2 · GUI 必要
交互原语
click · 指向与点击 type · 文字与表单 geom · 精确几何
工作产物
WP09 · 文件与文件夹
条件
licensed · 授权专业软件
提案难度
3 / 5

建议验证方式

lrcat GPS for each photo vs GT positions interpolated with the true offset: median error ≤ 30 m, ≥ 95% within 100 m; capture times may remain unchanged. Photos outside the track time span must stay untagged.

建议部分得分

30% offset identified; 50% position accuracy; 20% out-of-span photos untagged.

R06.02.01.04-T12 Cull and species-tag a 300-frame bird shoot in digiKam

原始指令

digiKam album fixtures/wetland/ has 300 bird photos. Rate 1 star every frame with motion blur or missed focus on the bird, 3–5 stars the rest by quality, and assign species tags from input/field_list.csv (12 species) using the reference sheet input/species_ref.pdf.

平台
lin · Linux mac · macOS win · Windows
应用
digiKam
交互表面
desk · 桌面 GUI
GUI 需求 · 任务设定
G2 · GUI 必要
GUI 需求 · 普通权限
G2 · GUI 必要
交互原语
click · 指向与点击 vis · 视觉判断 nav · 平移 / 缩放 / 滚动
工作产物
WP09 · 文件与文件夹
条件
long · 长流程 tutorial · 阅读文档 / 教程
提案难度
5 / 5

建议验证方式

digikam4.db: 1-star set vs GT technical rejects (F1 ≥ 0.85); species tags per image vs GT labels (accuracy ≥ 90%); no image with > 1 species unless GT has multiple. Species set restricted to the field list. exiftool/exiv2 CLIs are not installed on the VM and digikam4.db may only be written by the digiKam process; sqlite3 edits are logged and rejected.

建议部分得分

35% technical cull F1; 45% species accuracy; 20% ratings complete for all 300.

R06.02.01.04-T13 Caption 60 match photos with player names by visible jersey

原始指令

Photo Mechanic contact sheet fixtures/match/ has 60 frames. Caption each with the template '{Name} (#{num}) of {Team} during the first half' for every player whose number is clearly visible, using code replacements from input/roster.txt; frames with no readable number get 'Action during the first half'.

平台
mac · macOS win · Windows
应用
Photo Mechanic
交互表面
desk · 桌面 GUI
GUI 需求 · 任务设定
G2 · GUI 必要
GUI 需求 · 普通权限
G2 · GUI 必要
交互原语
type · 文字与表单 vis · 视觉判断 nav · 平移 / 缩放 / 滚动
工作产物
WP09 · 文件与文件夹
条件
licensed · 授权专业软件
提案难度
4 / 5

建议验证方式

IPTC Caption-Abstract per file vs GT visible-number lists: per-frame exact name set match; precision and recall over all player mentions ≥ 0.9 for full credit; template wording exact. exiftool/exiv2 are not installed on the VM and writes to fixtures/match/ are logged by process, so captions must be written by Photo Mechanic.

建议部分得分

60% player mentions F1; 25% template wording; 15% fallback captions correct.

R06.02.01.04-T14 Synchronize capture times of a second camera from a clock photo

原始指令

Lightroom Classic catalog fixtures/twocam.lrcat has images from camera A (correct time) and camera B (unknown offset). Frame B_0001 shows a wall clock also captured as A_0003. Fix camera B capture times with Edit Capture Time so both cameras interleave correctly, then create collection 'Merged' sorted by capture time.

平台
mac · macOS win · Windows
应用
Lightroom Classic
交互表面
desk · 桌面 GUI
GUI 需求 · 任务设定
HY · 混合
GUI 需求 · 普通权限
G2 · GUI 必要
交互原语
click · 指向与点击 type · 文字与表单 vis · 视觉判断
工作产物
WP09 · 文件与文件夹
条件
licensed · 授权专业软件
提案难度
3 / 5

建议验证方式

lrcat captureTime for all B images shifted by the GT offset ± 2 s; A images unchanged; 'Merged' collection contains all images. Interleaving order matches GT event order for 20 key moments.

建议部分得分

40% correct offset; 20% A untouched; 25% interleaving order; 15% collection created.

R06.02.01.04-T15 Pick a 40-image story edit from a 900-frame wedding day

原始指令

Lightroom Classic catalog fixtures/fullday.lrcat (900 frames) needs an album edit of exactly 40 images in collection 'Album_40' following input/story_brief.md: required beats (getting ready, vows, rings, first kiss, family formal, first dance, exit), no near-duplicates, no technical rejects, chronological order via custom sort.

平台
mac · macOS win · Windows
应用
Lightroom Classic
交互表面
desk · 桌面 GUI
GUI 需求 · 任务设定
G2 · GUI 必要
GUI 需求 · 普通权限
G2 · GUI 必要
交互原语
click · 指向与点击 drag · 拖放 vis · 视觉判断 nav · 平移 / 缩放 / 滚动
工作产物
WP09 · 文件与文件夹
条件
licensed · 授权专业软件 long · 长流程
提案难度
5 / 5

建议验证方式

lrcat: 'Album_40' has 40 images; each GT beat covered by ≥ 1 image from that beat's GT set; no two from the same near-duplicate cluster; none from GT technical-reject set; custom order monotonic in capture time. VLM rubric not used. Lightroom's Lua plug-ins are disabled and the .lrcat may only be written by the Lightroom process; sqlite3 edits from other processes are logged and rejected.

建议部分得分

20% count = 40 in chronological order; 40% beat coverage (7 beats); 20% no duplicates; 20% no technical rejects.

R06.02.01.04-T16 Batch-rename and copyright-stamp a folder in Bridge

原始指令

In Adobe Bridge, rename the 200 files in fixtures/studio_0310/ to Studio_0310_{seq3}.{ext} preserving capture order, keep the original filename in XMP, and append copyright '© 2026 Lumen Studio' plus usage terms 'Editorial use only' to all files.

平台
mac · macOS win · Windows
应用
Adobe Bridge
交互表面
desk · 桌面 GUI
GUI 需求 · 任务设定
HY · 混合
GUI 需求 · 普通权限
CL · CLI 足够
交互原语
click · 指向与点击 type · 文字与表单
工作产物
WP09 · 文件与文件夹
条件
licensed · 授权专业软件
提案难度
2 / 5

建议验证方式

200 files named per pattern in DateTimeOriginal order; XMP xmpMM:PreservedFileName equals old name; dc:rights and xmpRights:UsageTerms exact on all files; sidecars renamed with raws.

建议部分得分

40% names and order; 20% preserved filenames; 30% rights fields; 10% sidecars consistent.

R06.02.01.04-T17 Control: apply the T02 IPTC fields with exiftool

原始指令

Control pair of T02 (CLI half). In a terminal, write the same four IPTC/XMP fields as T02 (creator, copyright, city Lyon, hierarchical keyword Event|Gala) to the 118 B_*.CR3 sidecars in fixtures/Event_0914/ with exiftool, leaving A_* files untouched.

平台
lin · Linux mac · macOS win · Windows
应用
exiftool
交互表面
term · 终端
GUI 需求 · 任务设定
CL · CLI 足够
GUI 需求 · 普通权限
CL · CLI 足够
交互原语
code · 代码与命令 type · 文字与表单
工作产物
WP09 · 文件与文件夹
条件
未声明
提案难度
2 / 5

建议验证方式

Read XMP sidecars: all B_* have the four fields exactly; no A_* sidecar modified (hash unchanged). Pairs with T02 to measure the GUI-catalog route against the file route.

建议部分得分

60% B files correct; 30% A untouched; 10% hierarchical keyword format.

R06.02.01.04-T18 Choose the best variant per room in a real-estate shoot

原始指令

Lightroom Classic folder 'Elm_St' has 75 photos covering 16 rooms (3–6 variants each). For each room pick the one with straight verticals, no photographer/flash reflection and best window exposure; green-label it and add it to publish collection 'MLS_Upload' in room order from input/room_order.txt.

平台
mac · macOS win · Windows
应用
Lightroom Classic
交互表面
desk · 桌面 GUI
GUI 需求 · 任务设定
G2 · GUI 必要
GUI 需求 · 普通权限
G2 · GUI 必要
交互原语
click · 指向与点击 drag · 拖放 vis · 视觉判断
工作产物
WP09 · 文件与文件夹
条件
licensed · 授权专业软件
提案难度
4 / 5

建议验证方式

GT acceptable keepers per room (1–2 each, defined by vertical tilt ≤ 1°, no reflection, window clipping ≤ 2%). lrcat: exactly 16 green labels, one per room, all acceptable; publish collection order matches room order. Lightroom's Lua plug-ins are disabled and the .lrcat may only be written by the Lightroom process; sqlite3 edits from other processes are logged and rejected.

建议部分得分

60% acceptable keeper per room (scaled); 20% one per room; 20% collection and order.

R06.02.01.04-T19 Flag images that need model releases

原始指令

Lightroom Classic catalog fixtures/street_fest.lrcat holds 220 synthetic street-festival photos. Add keyword 'NeedsRelease' to every image where a person's face is identifiable (frontal or 3/4, ≥ 80 px face height) and 'NoRelease' to the rest; build smart collection 'Commercial_OK' = NoRelease.

平台
mac · macOS win · Windows
应用
Lightroom Classic
交互表面
desk · 桌面 GUI
GUI 需求 · 任务设定
G2 · GUI 必要
GUI 需求 · 普通权限
G2 · GUI 必要
交互原语
click · 指向与点击 key · 快捷键 vis · 视觉判断
工作产物
WP09 · 文件与文件夹
条件
licensed · 授权专业软件
提案难度
3 / 5

建议验证方式

lrcat keywords vs GT face annotations (identifiable per the stated rule): accuracy ≥ 95% for full credit, every image has exactly one of the two keywords; smart collection rule resolves to NoRelease set. Lightroom's Lua plug-ins are disabled and the .lrcat may only be written by the Lightroom process; sqlite3 edits from other processes are logged and rejected.

建议部分得分

60% classification accuracy; 20% every image labeled once; 20% smart collection.

R06.02.01.04-T20 Merge three Capture One catalogs and clean the keyword tree

原始指令

Merge fixtures/cat_2023, cat_2024, cat_2025 Capture One catalogs into out/Archive.cocatalog. Relink the 2024 folder that moved to /Volumes/Archive/2024, merge case and spelling variants of keywords per input/keyword_map.csv into one hierarchy, and remove empty keywords.

平台
mac · macOS win · Windows
应用
Capture One
交互表面
desk · 桌面 GUI
GUI 需求 · 任务设定
HY · 混合
GUI 需求 · 普通权限
G2 · GUI 必要
交互原语
click · 指向与点击 drag · 拖放 type · 文字与表单
工作产物
WP09 · 文件与文件夹
条件
licensed · 授权专业软件 long · 长流程
提案难度
4 / 5

建议验证方式

Catalog DB: image count = sum of three (no duplicates); 0 offline images; keyword list equals the mapped canonical set; each image's keywords mapped correctly (≥ 99%); no empty keywords.

建议部分得分

25% merged count; 25% relinked offline folder; 35% keyword consolidation; 15% empty keywords removed.

已有研究映射

当前映射未给此小类指定基准参考。

同工作流族的小类