Skip to content

Releases: nyabi-gh/CloakFrame

v1.11.2

Choose a tag to compare

@github-actions github-actions released this 14 Sep 11:59

작업이 끝난 뒤 어떤 파일을 다시 봐야 하는지 문제 유형별로 찾아갈 수 있고,
영상에서 추적이 끊긴 구간을 순서대로 확인하며 기록을 남길 수 있습니다. 파일을
추가할 때 창이 멈추지 않고, 로컬 로그에는 파일명을 남기지 않습니다.

파일별 결과

  • 작업이 끝나면 활동 영역의 **파일별 결과…**에서 확인 필요·실패 항목을 문제
    유형별로 걸러 볼 수 있습니다. 미처리 영역, 추적 공백, 자동·사용자 제외 트랙,
    스캔 실패, 메타데이터 경고, 출력 충돌, 처리 실패, 아무것도 가리지 않은 출력을
    나눠서 봅니다. 이전에는 활동 로그를 직접 훑어 문제 파일을 찾아야 했습니다.
  • 목록에서 입력 파일을 열거나, 선택한 입력을 다시 검토하거나 재처리할 수
    있습니다. 공백을 선택하면 다시 분석한 뒤 그 프레임에서 영상 검토를 시작합니다.
  • 재처리는 현재 설정으로 기존 출력 아래 새 작업 폴더에 저장합니다. 원본을
    처음부터 다시 분석하므로 이전 편집은 남지 않습니다. 버튼에도 같은 안내가
    있습니다.
  • 출력 폴더 열기는 실제로 저장된 결과에만 쓸 수 있습니다. 결과는 다음 작업을
    시작할 때 초기화되며, 아직 시작하지 않은 파일은 목록에 넣지 않습니다.
  • 병렬 처리를 취소해도 이미 실행 중이던 파일의 결과를 빠뜨리지 않습니다.

영상 검토

  • 검토 전에 발견한 추적 공백을 시작 프레임 순으로 나열합니다. 각 항목은 트랙
    번호와 밀리초 단위 시간, 1부터 시작하는 프레임 범위를 보여 주고, 선택하면 그
    구간의 첫 프레임으로 이동합니다.
  • 이전 공백 / 다음 공백은 현재 위치 앞뒤의 가장 가까운 시작점으로
    이동합니다. 시작점이 같은 공백은 버튼으로 한 번만 지나가며, 목록에서는 각각
    선택할 수 있습니다.
  • 좌우 방향키나 타임라인 옆 버튼으로 한 프레임씩 이동합니다. 프레임을 바꾸면
    그리던 박스를 취소해 다른 프레임에 잘못 적용되지 않게 합니다.
  • 공백을 모두 살펴본 뒤 확인 완료를 체크하면 검토 기록으로 남고, 아직
    확인하지 않은 프레임 수를 따로 볼 수 있습니다. 마스크를 추가·수정·삭제하거나
    트랙 포함 여부를 바꾸면 체크는 초기화됩니다. 확인 완료는 사용자가 봤다는
    기록이며, 그 구간이 실제로 가려졌는지 앱이 검증했다는 뜻이 아닙니다.
  • 타임라인의 빨간 표시와 경고는 검토 전에 발견한 추적 공백을 뜻합니다.
    수동 마스크를 얹는 것만으로 대상의 실제 위치가 확인되지는 않으므로 경고는
    그대로 남습니다. 공유하기 전에 해당 구간을 직접 재생해 확인해 주세요.
  • 저신뢰 트랙 안내를 실제 동작인 기본 포함에 맞게 고쳤습니다. 이전에는 설명만
    기본 제외라고 되어 있어 목록의 체크 상태와 어긋났습니다.

결과 요약과 읽지 못한 입력

  • 요약에서 검출 제한으로 제외된 영역, 추적 공백 프레임, 제외된 트랙, 메타데이터
    경고가 있는 파일, 읽지 못한 입력 경로를 구분해 표시합니다. 이전에는 하나의
    수치로 합쳐져 무엇이 문제인지 알기 어려웠습니다.
  • 접근할 수 없는 하위 폴더를 정확한 경로와 함께 알리고, 읽을 수 있는 다른
    폴더는 계속 훑습니다. 디렉터리 심볼릭 링크는 재귀적으로 따라가지 않습니다.

파일을 추가할 때

  • 사진과 영상 썸네일을 백그라운드에서 최대 2개씩 만듭니다. 목록에는 기본
    아이콘이 먼저 나타나고 준비되는 대로 채워집니다. 이전에는 영상을 추가할
    때마다 창이 멈춘 것처럼 보일 수 있었습니다.
  • 목록에서 지운 항목의 썸네일 결과는 버립니다. 아주 크거나 크기를 알 수 없는
    이미지는 기본 아이콘을 그대로 둡니다. 썸네일을 만들지 않는 것과 본 처리의
    제한은 별개입니다.

로컬 로그

  • 기본 디스크 로그에 파일명과 경로를 남기지 않습니다. 시작 여부와 결과 수치 같은
    진단 정보만 기록합니다.
  • 파일명이 들어갈 수 있는 기록은 상세 로그를 직접 켜야 남습니다. 기본값은
    꺼짐이고, 이전에 파일 로그를 켜 두었더라도 상세 기록에 동의한 것으로
    넘겨받지 않습니다.
  • 로그 설정은 즉시 적용되며, 설정에서 로그 폴더를 열거나 로그를 지울 수
    있습니다. 삭제는 로그 파일과 순환 로그만 지웁니다. 화면의 활동 내역은 이 디스크
    로그 정책과 별개입니다.

모델 다운로드

  • 30초 동안 응답이 없거나, 재시도 대기를 포함해 15분을 넘기면 중단합니다.
    일시적인 연결 실패와 HTTP 429·5xx, 시간 초과는 1초 뒤 다시 시도하며 최대
    3회까지 받습니다. 취소는 내려받는 중에도 재시도를 기다리는 중에도 반영됩니다.
  • 취소·시간 초과·네트워크 실패·크기 초과·무결성 실패를 구분해 알립니다. 크기와
    무결성 실패는 다시 시도하지 않으며, SHA-256 검증을 통과한 데이터만 저장합니다.
English

Lets you find the files that still need a look by the kind of problem they
have, walk a video's tracking gaps in order and record the ones you checked,
and keeps the window responsive while files are added. Local logs no longer
carry file names.

File results

  • When a run finishes, File results… in the activity area filters the
    attention and failure items by the kind of problem: uncovered regions,
    tracking gaps, tracks dropped automatically or by you, scan failures,
    metadata warnings, output conflicts, processing failures, and output that
    covered nothing. Until now you had to read the activity log to find which
    file went wrong.
  • Open an input from the list, or review or reprocess a selected input. Picking
    a gap re-analyses the video and opens review at that frame.
  • Reprocessing uses your current settings and writes to a new job folder
    under the existing output. It analyses the original from scratch, so earlier
    edits are not kept. The button says the same thing.
  • Opening the output folder works only for results that were actually saved.
    Results reset when the next run starts, and files that never started are not
    listed.
  • Cancelling a parallel run no longer drops the results of files that were
    already being processed.

Video review

  • Tracking gaps found before review are listed in start-frame order. Each entry
    shows the track number, the time in milliseconds, and a frame range counted
    from 1; selecting one jumps to the first frame of that gap.
  • Previous gap / Next gap move to the nearest start before or after the
    current position. Gaps that share a start are passed once by the buttons and
    can still be selected individually in the list.
  • Left and Right arrow keys, or the buttons beside the timeline, move one frame
    at a time. Changing frames cancels a box you were drawing so it cannot land
    on the wrong frame.
  • After looking through a gap you can tick Reviewed, which records that you
    checked it and lets you see how many frames are still unchecked. Adding,
    editing, or deleting a mask, or changing whether a track is included, clears
    the ticks. Reviewed is a record of what you looked at, not the app verifying
    that the range ended up covered.
  • The red timeline marks and the warnings mean tracking gaps found before
    review
    . Adding a manual mask does not establish where the subject actually
    was, so the warnings stay. Play those ranges yourself before sharing.
  • The low-confidence track text now matches what the app does: they are
    included by default. The description said they were excluded, which
    contradicted the checkboxes in the list.

Result summary and inputs that could not be read

  • The summary now separates regions dropped at the detection limit, tracking
    gap frames, dropped tracks, files with metadata warnings, and input paths
    that could not be read. They used to be added into one number that did not
    say what had gone wrong.
  • A subfolder that cannot be opened is reported with its exact path, and the
    other readable folders are still scanned. Directory symlinks are not followed
    recursively.

Adding files

  • Photo and video thumbnails are built in the background, at most two at a
    time. The list shows a default icon first and fills in as they are ready.
    Adding a video used to make the window look frozen.
  • Thumbnails for entries you removed are discarded. Very large images, and
    images whose size cannot be read, keep the default icon. Skipping a thumbnail
    is not a limit on the processing itself.

Local logs

  • The basic disk log no longer records file names or paths. It keeps
    diagnostics such as whether a run started and how many files ended each way.
  • Anything that may contain a file name is written only when you turn on
    Detailed logging yourself. It is off by default, and an earlier
    fileLogging setting is not carried over as consent to detailed records.
  • Log settings apply immediately, and Settings can open the log folder or
    delete the logs. Deleting removes the log file and its rotations only. The
    on-screen activity list is separate from this disk-log policy.

Model downloads

  • A download stops after 30 seconds without a response, or after 15 minutes
    overall including the wait between attempts. A dropped connection, HTTP 429
    or 5xx, and a timeout are retried after one second, up to three attempts in
    total. Cancelling is honoured both while downloading and while waiting to
    retry.
  • Cancellation, timeout, network failure, oversize, and integrity failure are
    reported as distinct outcomes. Oversize and integrity failures are not
    retried, and only data that passes the SHA-256 check is stored.
日本語

処理のあとで見直すべきファイルを問題の種類ごとにたどれるようにし、動画で追跡が
途切れた区間を順に確認して記録を残せるようにしました。ファイルを追加しても画面が
固まらず、ローカルログにファイル名を残しません。

ファイル別の結果

  • 処理が終わると、活動エリアの ファイル別の結果… で要確認・失敗の項目を
    問題の種類ごとに絞り込めます。未処理の領域、追跡の空白、自動またはご自身で
    除外したトラック、スキャンの失敗、メタデータ警告、出力の衝突、処理の失敗、
    何も覆わなかった出力を分けて表示します。これまでは活動ログを自分でたどって
    問題のファイルを探す必要がありました。
  • 一覧から入力ファイルを開いたり、選んだ入力をもう一度確認したり再処理したり
    できます。空白を選ぶと、再解析したうえでそのフレームから動画の確認を始めます。
  • 再処理は 現在の設定 を使い、既存の出力の下に新しいジョブフォルダーを作って
    保存します。元のファイルを一から解析し直すため、以前の編集は残りません。
    ボタンにも同じ説明があります。
  • 出力フォルダーを開く操作は、実際に保存された結果でのみ使えます。結果は次の
    処理を始めるときにリセットされ、まだ始まっていないファイルは一覧に入りません。
  • 並列処理をキャンセルしても、すでに実行中だったファイルの結果を取りこぼしません。

動画の確認

  • 確認の前に見つかった追跡の空白を、開始フレーム順に並べます。各項目はトラック
    番号、ミリ秒単位の時刻、1から数えたフレーム範囲を示し、選ぶとその区間の最初の
    フレームへ移動します。
  • 前の空白 / 次の空白 は、現在の位置より前後で最も近い開始点へ移動します。
    開始点が同じ空白はボタンでは一度だけ通過し、一覧ではそれぞれ選べます。
  • 左右の矢印キー、またはタイムライン横のボタンで1フレームずつ移動します。
    フレームを変えると描画中の枠を取り消し、別のフレームに誤って適用されるのを
    防ぎます。
  • 空白をひととおり見たあとに 確認済み をチェックすると確認の記録として残り、
    まだ確認していないフレーム数を別に見られます。マスクの追加・編集・削除や、
    トラックを含めるかどうかの変更を行うと、チェックはリセットされます。確認済みは
    ご自身が見たという記録であり、その区間が実際に覆われたことをアプリが検証した
    という意味ではありません。
  • タイムラインの赤い印と警告は、確認の前に見つかった 追跡の空白を表します。
    手動マスクを足しても対象が実際にどこにいたかは確かめられないため、警告は残り
    ます。共有する前に、その区間をご自身で再生して確認してください。
  • 低信頼度トラックの説明を、実際の動作である既定で含めるに合わせました。これまで
    説明だけが既定で除外となっており、一覧のチェック状態と食い違っていました。

結果の概要と読み取れなかった入力

  • 概要では、検出の上限で除外された領域、追跡の空白フレーム、除外されたトラック、
    メタデータ警告のあるファイル、読み取れなかった入力パスを分けて表示します。
    これまでは一つの数値にまとめられ、何が問題なのか分かりませんでした。
  • 開けないサブフォルダーは正確なパスとともに知らせ、読み取れるほかのフォルダーの
    スキャンは続けます。ディレクトリのシンボリックリンクは再帰的にたどりません。

ファイルを追加するとき

  • 写真と動画のサムネイルを、最大2件ずつバックグラウンドで作ります。一覧には
    まず既定のアイコンが出て、できあがった順に置き換わります。これまでは動画を
    追加するたびに画面が固まったように見えることがありました。
  • 一覧から削除した項目のサムネイルは破棄します。とても大きい画像や、サイズが
    読み取れない画像は既定のアイコンのままにします。サムネイルを作らないことと、
    処理そのものの制限は別です。

ローカルログ

  • 基本のディスクログにファイル名やパスを残しません。処理が始まったかどうかや
    結果の件数といった診断情報だけを記録します。
  • ファイル名が含まれうる記録は、詳細ログ をご自身で有効にしたときだけ残り
    ます。既定は無効で、以前のファイルログの設定を詳細な記録への同意として
    引き継ぐことはしません。
  • ログの設定はすぐに反映され、設定からログフォルダーを開いたりログを削除したり
    できます。削除はログファイルとそのローテーションのみを消します。画面上の活動
    内容は、このディスクログの方針とは別です。

モデルのダウンロード

  • 30秒応答がないとき、または再試行の待ち時間を含めて全体で15分を超えたときは
    中止します。一時的な接続の失敗、HTTP 429・5xx、タイムアウトは1秒後に再試行し、
    合計3回まで試します。キャンセルは、ダウンロード中も再試行の待機中も反映され
    ます。
  • キャンセル・タイムアウト・ネットワークの失敗・サイズ超過・整合性エラーを
    区別して知らせます。サイズと整合性の失敗は再試行せず、SHA-256の検証を通った
    データだけを保存します。
简体中文

让您在处理结束后按问题类型找到仍需查看的文件,按顺序检查视频中跟踪中断的区间
并记录已看过的部分。添加文件时窗口不再卡顿,本地日志也不再保留文件名。

文件结果

  • 处理结束后,可在活动区域的 文件结果… 中按问题类型筛选需要关注和失败的
    项目:未遮盖的区域、跟踪空白、自动或由您排除的轨迹、扫描失败、元数据警告、
    输出冲突、处理失败,以及什么都没有遮盖的输出。以前您需要自己翻阅活动日志
    才能找出出问题的文件。
  • 可从列表中打开输入文件,或对选中的输入重新检查、重新处理。选择某个空白后,
    程序会重新分析并从该帧开始视频检查。
  • 重新处理使用 当前设置,并在现有输出下新建作业文件夹保存。它会从头重新
    分析原始文件,因此不会保留之前的编辑。按钮上也有同样的说明。
  • 只有实际保存了结果时才能打开输出文件夹。开始下一次处理时结果会重置,尚未
    开始的文件不会列出。
  • 取消并行处理时,不再丢失已经开始处理的文件的结果。

视频检查

...

Read more

v1.11.1

Choose a tag to compare

@github-actions github-actions released this 24 Aug 02:23

검토를 건너뛴 결과나 아무것도 가리지 않은 결과가 조용히 저장되지 않도록 하고,
읽지 못한 입력을 알려 주며, 영상과 사진 일괄 처리가 빨라집니다.

저장 전 검토

  • 새로 설치하면 저장 전 검토가 켜진 상태로 시작합니다. 공유하기 전에 결과를
    확인하도록 권해 왔는데 정작 처음 실행할 때는 꺼져 있었습니다. 필요 없으면
    그대로 끄면 되고, 이미 쓰던 설정은 바뀌지 않습니다.
  • 검토를 켰는데 검토 화면을 띄울 수 없으면 이제 아무것도 저장하지 않고 실패로
    알립니다. 이전에는 검토를 건너뛴 채 검출한 그대로 저장했고, 그러고도 작업이
    완료로 끝날 수 있었습니다. 사진과 영상 모두 해당합니다.
  • 검출된 영역을 모두 제외하고 대신 추가한 것도 없으면, 저장하기 전에 한 번 더
    묻습니다. 자동 검출이 처음부터 0개였던 경우는 묻지 않습니다. 얼굴이 없는
    사진은 그대로 저장되는 것이 맞고, 그때마다 물으면 정작 중요한 확인을 그냥
    넘기게 됩니다. 영상도 같습니다.

읽지 못한 입력

  • 폴더를 훑다가 읽을 수 없는 파일이나 폴더를 만나면 어느 것이 왜 안 됐는지
    알리고, 작업은 완료가 아니라 검토 필요로 끝납니다. 이전에는 조용히
    빠졌고, 오류 하나로 남은 폴더를 아예 훑지 않고 멈추는 일도 있었습니다.
  • 이미 없는 경로도 그냥 넘어가지 않고 알려 줍니다.

처리 속도

  • 영상 처리가 빨라졌습니다. 이전에는 디코딩이 검출·인코딩과 번갈아 가며
    진행됐지만, 이제 나란히 진행되어 서로 기다리는 시간이 줄었습니다. 디코딩이
    무거운 영상일수록 효과가 큽니다.
  • 여러 장의 사진을 한꺼번에 처리할 때 더 빨리 끝납니다. 사진들이 검출 차례를
    기다리지 않고, 번호판 검출도 얼굴 검출을 기다리지 않습니다.
  • macOS에서는 검출 모델을 매번 새로 준비하는 대신 한 번 준비해 기억해 두므로,
    두 번째 실행부터는 작업이 더 빨리 시작됩니다.
  • GPU 가속 없이 검출하는 컴퓨터에서도 검출이 조금 빨라졌습니다.

업데이트

  • 공식 Windows·Linux 빌드는 이제 업데이트 서명 키 없이는 배포될 수 없습니다.

표시 언어

  • 일본어에서 일부가 영어로 남아 있던 메시지를 고쳤습니다. 건너뛴 파일, 복사·저장
    실패, FFmpeg 오류처럼 작업이 무엇을 하지 못했는지 알리는 메시지들입니다.
English

Makes sure nothing is saved without the review it promised or with nothing
covered, tells you about inputs it could not read, and processes videos and
photo batches faster.

Review before saving

  • A fresh install now starts with Review before saving on. The app has
    always asked you to check the result before sharing it, yet the very first
    run started with the check turned off. Turn it off if you do not want it, and
    a setting you have already chosen is left alone.
  • If review is on and the review window cannot be shown, nothing is saved and
    the file is reported as failed. Until now the review was skipped, the
    detections were saved exactly as found, and the run could still end in
    Done. This applies to photos and videos alike.
  • If every region the detector found is excluded and nothing is added in its
    place, you are asked once more before the file is saved. A run that detected
    nothing to begin with is not asked about: a photo with no face in it should
    simply be saved, and prompting every time would teach you to click through
    the one prompt that matters. Video works the same way.

Inputs that could not be read

  • When a folder scan meets a file or folder it cannot read, you are told which
    one and why, and the run ends in Review required rather than Done.
    Until now they dropped out in silence, and a single error could stop the rest
    of the folder from being scanned at all.
  • A path that is no longer there is reported instead of passed over.

Faster processing

  • Videos are processed faster. Decoding used to take turns with detection and
    with encoding; now they run side by side, so less time is spent waiting on
    one another. Videos that are slower to decode gain the most.
  • A batch of photos finishes sooner: photos no longer wait for one another
    before detection, and license-plate detection no longer waits for face
    detection.
  • On macOS the detection models are prepared once and remembered instead of
    being rebuilt every time the app starts, so from the second launch on, a run
    begins sooner.
  • Detection is a little faster on computers that run it without a supported
    GPU.

Updates

  • An official Windows or Linux build can no longer be published without an
    update signing key.

Language

  • Japanese messages that were still partly in English are fixed. They are the
    lines that say what a run could not do: files skipped, a copy or save that
    failed, FFmpeg errors.
日本語

確認を飛ばした結果や何も覆っていない結果が黙って保存されないようにし、
読み取れなかった入力を知らせ、動画と写真の一括処理を速くします。

保存前に確認

  • 新しくインストールすると 保存前に確認 が有効な状態で始まります。共有する
    前に結果を確かめるようお願いしてきたのに、最初の実行では確認が切れていました。
    不要ならそのままオフにでき、すでに選んである設定は変わりません。
  • 確認が有効なのに確認画面を表示できないときは、何も保存せず失敗として知らせ
    ます。これまでは確認を飛ばして検出したまま保存し、それでも処理が 完了
    終わることがありました。写真も動画も同じです。
  • 検出された領域をすべて除外し、代わりに何も追加していない場合は、保存する前に
    もう一度確認します。最初から検出が 0 件だった場合は確認しません。顔のない
    写真はそのまま保存されるべきですし、毎回確認していては肝心な確認まで
    読み飛ばすようになってしまいます。動画も同じです。

読み取れなかった入力

  • フォルダーをたどる途中で読み取れないファイルやフォルダーに出会うと、どれが
    なぜ読めなかったかを知らせ、処理は 完了 ではなく 確認が必要です
    終わります。これまでは黙って抜け落ち、一度のエラーで残りのフォルダーを
    まったくたどらずに止まることもありました。
  • すでに存在しないパスも、素通りせずに知らせます。

処理速度

  • 動画の処理が速くなりました。これまではデコードが検出やエンコードと交互に
    進んでいましたが、いまは並行して進むため、互いを待つ時間が減りました。
    デコードの重い動画ほど効果が大きくなります。
  • 複数の写真をまとめて処理するときも早く終わります。写真が検出の順番待ちを
    しなくなり、ナンバープレートの検出も顔の検出を待ちません。
  • macOS では検出モデルを毎回作り直す代わりに一度準備して覚えておくため、
    2 回目の起動からは処理がより早く始まります。
  • GPU による高速化なしで検出するコンピューターでも、検出が少し速くなりました。

アップデート

  • 公式の Windows・Linux ビルドは、アップデートの署名鍵なしでは公開できなく
    なりました。

表示言語

  • 一部が英語のままだった日本語のメッセージを修正しました。スキップしたファイル、
    コピーや保存の失敗、FFmpeg のエラーなど、処理が何をできなかったかを伝える
    メッセージです。
简体中文

确保不会悄悄保存未经检查或未遮挡任何内容的结果,告知无法读取的输入,并加快
视频与照片批量处理的速度。

保存前检查

  • 全新安装现在会默认开启 保存前检查。应用一直建议你在分享前确认结果,可首次
    运行时这项检查却是关闭的。不需要可以随时关掉,你已经设定过的选项不会被改动。
  • 开启检查后如果无法显示检查窗口,现在不会保存任何内容,而是把该文件报告为失败。
    此前会跳过检查、按检测结果直接保存,而且任务仍可能以 完成 结束。照片和
    视频都是如此。
  • 如果检测到的区域全部被排除,又没有添加任何区域来替代,保存前会再确认一次。
    一开始就没有检测到任何内容的情况不会询问:没有人脸的照片本就应该直接保存,
    每次都询问反而会让你对真正重要的提示视而不见。视频同理。

无法读取的输入

  • 扫描文件夹时遇到无法读取的文件或文件夹,现在会告诉你是哪一个、为什么读不了,
    任务也会以 需要检查 而不是 完成 结束。此前它们会悄无声息地消失,而且
    一次错误就可能导致文件夹的其余部分完全不被扫描。
  • 已经不存在的路径也会被报告,而不是直接略过。

处理速度

  • 视频处理更快了。以前解码要与检测、编码轮流进行,现在它们并行进行,相互等待
    的时间随之减少。解码越慢的视频提升越明显。
  • 批量处理多张照片时完成得更快:照片不再排队等待检测,车牌检测也不再等待
    人脸检测。
  • 在 macOS 上,检测模型现在只准备一次并记住结果,而不是每次启动都重新构建,
    因此从第二次启动起任务开始得更快。
  • 在没有 GPU 加速的电脑上,检测也稍微快了一些。

更新

  • 官方的 Windows 和 Linux 版本现在没有更新签名密钥就无法发布。

界面语言

  • 修复了部分仍为英文的日语消息。它们正是说明任务未能完成什么的提示:被跳过的
    文件、复制或保存失败、FFmpeg 错误。

v1.11.0

Choose a tag to compare

@github-actions github-actions released this 11 Aug 02:46

업데이트

  • 업데이트를 수락하면 앱이 알아서 설치합니다. Windows와 Linux는 그 자리에서
    갱신되고, macOS는 다운로드 페이지를 여는 대신 업데이트를 설치합니다.
  • 먼저 물어보는 것은 그대로입니다. 시작할 때의 확인은 설정에서 끌 수 있습니다.

새로운 얼굴 모델

  • 사진과 영상에는 YOLO5Face-n을 권장합니다.
  • YuNet은 느린 컴퓨터를 위한 가벼운 선택지로, 0.3 MB 미만으로 내려받습니다.
  • 찾아보기… 로 직접 준비한 ONNX 모델을 쓰는 것도 그대로 됩니다.

원하는 이미지로 얼굴 가리기

  • 모자이크 대신 쓸 PNG, JPEG, BMP, TIFF, WebP 이미지를 고를 수 있습니다. 사진,
    영상, 미리보기, 검토 화면에서 모두 적용됩니다.
  • 이미지는 원래 비율을 유지하고 얼굴이 기울면 같이 기울어집니다.
  • 영상에서는 얼굴이 움직이는 대로 부드럽게 따라갑니다.
  • 투명한 부분으로 원본이 비치는 일은 없습니다.

webm 영상

  • 기존 형식과 함께 .webm 파일도 넣을 수 있습니다.

영상 검토

  • 영상을 저장하기 전에, 앱이 놓친 부분을 직접 상자로 그릴 수 있습니다.
  • 상자마다 적용할 구간을 정할 수 있습니다.
  • 대상이 움직이면 상자를 옮겨 주기만 하면 그 사이 프레임은 알아서 채워집니다.
  • 직접 그린 상자를 지우거나, 앱이 잘못 잡은 것을 계속 빼 둘 수 있습니다.

못 가렸을 수 있는 곳을 알려 줍니다

  • 놓쳤을 가능성이 있으면 — 사람이 너무 많은 사진에서 한도를 넘긴 얼굴, 확신이
    부족해 빠진 대상, 앱이 채우지 못한 구간 — 작업이 완료가 아니라 검토
    필요
    로 끝나고 몇 개인지 알려 줍니다.
  • 뚜렷하게 잡힌 횟수가 적은 영상 대상도 이제 기본으로 포함합니다. 뺄지 말지는
    검토 화면에서 직접 정합니다.

영상

  • 검토 화면에서 본 그대로 저장되고, 소리도 화면과 어긋나지 않습니다.
  • 아나모픽 영상을 포함해 화면 비율이 올바르게 유지됩니다.
  • 소리 트랙을 모두 유지합니다. 이전에는 첫 번째만 남아서 해설이나 다른 언어
    트랙이 아무 말 없이 사라졌습니다.
  • 손상된 영상은 짧아진 결과를 조용히 저장하지 않고 실패로 알립니다.
  • 하드웨어 인코딩이 도중에 실패하면 파일을 포기하지 않고 소프트웨어로 마저
    끝냅니다.
  • 몇 프레임 사이에 장면이 두 번 바뀌어도 모자이크가 전환을 넘어 흘러가지
    않습니다.

안정성

  • 사용할 수 있는 메모리보다 큰 사진은 이유를 알리고 건너뜁니다. 이전에는 그대로
    진행하다 메모리가 부족해질 수 있었습니다.
  • 시작 직후에 누른 중지가 무시되지 않습니다.
  • 시스템 자원이 부족해도 앱이 갑자기 종료되지 않습니다.
  • 투명한 부분이 있는 이미지에서 단색으로 가릴 때 비쳐 보이지 않습니다.
  • 저장이 중간에 끊겨도 완성본처럼 보이는 반쪽짜리 파일이 남지 않습니다.
  • Windows에서 경로에 한글 등이 들어가도 사진 정보가 그대로 보존됩니다.
  • 기존의 빠름·정확 선택을 새 모델 목록으로 그대로 이어받습니다.
  • 내려받은 모델은 쓰기 전에 검사합니다.
  • 승인한 커스텀 ONNX 모델 파일이 바뀌면 불러오기 전에 다시 묻습니다.
  • 작업 도중 앱이 강제로 종료되거나 전원이 꺼지면 원본의 임시 사본이 그대로
    남았습니다. 이제 다음 실행 때 정리합니다.

표시 언어

  • 영상 오류와 진행 메시지가 한국어·일본어·중국어로 표시됩니다. 이전에는 어떤
    언어에서도 영어로 나왔습니다.

모델과 라이선스

  • 내장 모델은 처음 사용할 때 내려받은 뒤 오프라인에서 동작합니다.
  • YOLO5Face-n은 WIDER FACE 데이터로 학습해 비상업적 용도로만 취급해야 합니다.
  • YuNet은 MIT 라이선스이며, 직접 준비한 모델은 각자의 조건을 따릅니다.
English

Updates

  • Accepting an update now installs it for you. Windows and Linux update in
    place, and macOS installs the update instead of opening a download page.
  • You are still asked first, and the check at startup can be turned off in
    Settings.

New face models

  • YOLO5Face-n is the recommended model for photos and videos.
  • YuNet is a lighter choice for slower machines, under 0.3 MB to download.
  • You can still bring your own ONNX model through Browse….

Cover faces with your own image

  • Pick a PNG, JPEG, BMP, TIFF, or WebP image to use in place of a mosaic — in
    photos, videos, style previews, and review.
  • The image keeps its shape and tilts along with the face.
  • In video it follows the face smoothly as it moves.
  • Transparent parts never let the original show through.

webm videos

  • You can now drop .webm files in alongside the formats already supported.

Video review

  • Draw a box over anything the app missed, before the video is saved.
  • Set the stretch of time each box applies to.
  • Move a box as the subject moves, and the frames in between are filled in.
  • Remove boxes you drew, or keep leaving out ones the app found by mistake.

You are told what might not be covered

  • If anything could have been missed — faces beyond the limit in a crowded
    photo, a track dropped for being uncertain, a gap the app could not fill in —
    the run ends in Review required instead of Done and tells you how many.
  • Video tracks with only a few clear detections are now included by default.
    Leaving one out is your choice to make in review.

Video

  • What you see in review is exactly what gets saved, and the audio lines up
    with it.
  • Video keeps its correct proportions, anamorphic footage included.
  • Every audio track is kept. Before, only the first survived, so commentary and
    other-language tracks disappeared without a word.
  • A damaged video now fails instead of quietly saving a shortened result.
  • If hardware encoding fails partway through, the app finishes the job in
    software rather than giving up on the file.
  • Mosaics no longer drift across a cut when the shot changes twice within a few
    frames.

Reliability

  • A photo too large for the memory available is skipped with a reason, instead
    of risking the app running out.
  • Stop pressed right after Start is no longer ignored.
  • The app no longer quits when the system is short on resources.
  • Solid fill is no longer see-through on images with transparency.
  • An interrupted save no longer leaves a half-written file under the finished
    name.
  • On Windows, photo details are kept even when the path has non-English
    characters.
  • Your existing Fast or Accurate choice carries over to the new model list.
  • Downloaded models are checked before they are used.
  • If a custom ONNX model file changes after you approve it, you are asked again
    before it is loaded.
  • If the app is killed or the machine loses power mid-run, the temporary copy it
    made of your original is cleaned up the next time you start it. Until now it
    stayed behind.

Language

  • Video errors and progress messages now appear in Korean, Japanese, and
    Simplified Chinese. Until now they were in English whatever your language.

Models and licensing

  • The built-in models are downloaded the first time you use them, then work
    offline.
  • YOLO5Face-n is trained on WIDER FACE data and should be treated as
    non-commercial only.
  • YuNet is MIT-licensed, and models you bring yourself keep their own terms.
日本語

アップデート

  • アップデートを承認すると、アプリがそのままインストールします。Windows と
    Linux はその場で更新され、macOS はダウンロードページを開く代わりに
    アップデートをインストールします。
  • 先に確認する点は変わりません。起動時の確認は設定でオフにできます。

新しい顔モデル

  • 写真と動画には YOLO5Face-n をおすすめします。
  • YuNet は動作の遅いパソコン向けの軽い選択肢で、ダウンロードは 0.3 MB 未満です。
  • 参照… から自分で用意した ONNX モデルを使うこともこれまでどおりできます。

好きな画像で顔を覆う

  • モザイクの代わりに使う PNG、JPEG、BMP、TIFF、WebP の画像を選べます。写真・
    動画・スタイルのプレビュー・レビューのいずれでも使えます。
  • 画像は元の形を保ち、顔の傾きに合わせて一緒に傾きます。
  • 動画では顔の動きになめらかに追従します。
  • 透明な部分から元の映像が透けることはありません。

webm 動画

  • これまでの形式に加えて .webm ファイルも読み込めます。

動画レビュー

  • 動画を保存する前に、アプリが見逃した部分を自分で四角く囲めます。
  • 囲みごとに適用する時間の範囲を決められます。
  • 対象が動いたら囲みを動かすだけで、その間のコマは自動で埋まります。
  • 自分で描いた囲みを消したり、アプリが誤って拾ったものを外したままにできます。

隠しきれなかったかもしれない場所を知らせます

  • 見逃した可能性があるとき — 人が多すぎる写真で上限を超えた顔、確信が足りずに
    外れた対象、アプリが埋められなかった区間 — 処理は 完了 ではなく
    要確認 で終わり、その数を知らせます。
  • はっきり捉えられた回数が少ない動画の対象も、既定で含めるようになりました。
    外すかどうかはレビュー画面で自分で決めます。

動画

  • レビューで見たとおりに保存され、音もずれません。
  • アナモルフィックの映像も含めて、画面の比率が正しく保たれます。
  • 音声トラックをすべて残します。以前は最初の 1 本だけが残り、コメンタリーや
    別言語のトラックが何も言わずに消えていました。
  • 壊れた動画は、短くなった結果を黙って保存せず失敗として知らせます。
  • ハードウェアでのエンコードが途中で失敗しても、そのファイルを諦めずに
    ソフトウェアで最後まで処理します。
  • 数コマの間に場面が二度切り替わっても、モザイクが切り替わりを越えて流れません。

安定性

  • 使えるメモリより大きい写真は理由を知らせてスキップします。以前はそのまま進み、
    メモリ不足になるおそれがありました。
  • 開始 の直後に押した 停止 が無視されなくなりました。
  • システムの資源が足りないときでもアプリが突然終了しません。
  • 透明な部分がある画像を単色で覆っても透けなくなりました。
  • 保存が途中で止まっても、完成品に見える中途半端なファイルが残りません。
  • Windows でパスに日本語などが含まれていても写真の情報が保たれます。
  • これまでの「高速」「高精度」の選択が新しいモデル一覧にそのまま引き継がれます。
  • ダウンロードしたモデルは使う前に検査します。
  • 承認したカスタム ONNX モデルのファイルが変わっていた場合、読み込む前に改めて
    確認します。
  • 処理中にアプリが強制終了したり電源が切れたりすると、元のファイルの一時コピーが
    残ったままでした。次回の起動時に片づけます。

表示言語

  • 動画のエラーや進行状況のメッセージが日本語・韓国語・簡体字中国語で表示されます。
    これまではどの言語でも英語のままでした。

モデルとライセンス

  • 内蔵モデルは初回に一度ダウンロードすれば、あとはオフラインで動作します。
  • YOLO5Face-n は WIDER FACE のデータで学習しているため、非商用に限って扱って
    ください。
  • YuNet は MIT ライセンスで、自分で用意したモデルはそれぞれの条件に従います。
简体中文

更新

  • 同意更新后,应用会直接为你安装。Windows 和 Linux 就地更新,macOS 会安装更新,
    而不再打开下载页面。
  • 仍然会先征求你的同意,启动时的检查可以在设置中关闭。

新的人脸模型

  • 照片和视频推荐使用 YOLO5Face-n。
  • YuNet 是为较慢的电脑准备的轻量选择,下载不到 0.3 MB。
  • 仍然可以通过 浏览… 使用你自己准备的 ONNX 模型。

用自选图片遮住人脸

  • 可以选择 PNG、JPEG、BMP、TIFF、WebP 图片来代替马赛克,照片、视频、样式预览和
    审阅中都可使用。
  • 图片保持原有形状,并随人脸一起倾斜。
  • 在视频中会跟着人脸平滑移动。
  • 透明的部分不会透出原始画面。

webm 视频

  • 除已支持的格式外,现在也可以放入 .webm 文件。

视频审阅

  • 保存视频之前,可以给应用漏掉的地方自己画个框。
  • 可以为每个框设定生效的时间段。
  • 对象移动时只需跟着挪动框,中间的画面会自动补齐。
  • 可以删掉自己画的框,或继续排除应用认错的对象。

会告诉你哪里可能没遮住

  • 只要有可能遗漏 —— 人数过多的照片中超出上限的人脸、因为不够确定而被排除的对象、
    应用无法补齐的空档 —— 任务就会以 需要检查 而不是 完成 结束,并告诉你
    有多少处。
  • 清晰捕捉次数较少的视频对象现在默认也会包含在内,是否排除由你在审阅时决定。

视频

  • 审阅时看到的画面就是最终保存的画面,声音也不会错位。
  • 包括变形宽银幕在内,视频的画面比例都能正确保留。
  • 保留全部音频轨道。此前只有第一条能保留,解说和其他语言的轨道会悄无声息地丢失。
  • 损坏的视频不会再把变短的结果悄悄保存,而是报告失败。
  • 硬件编码中途失败时,应用不会放弃这个文件,而是改用软件把它做完。
  • 镜头在几帧之内连续切换两次时,马赛克也不会越过切换点继续漂移。

稳定性

  • 超出可用内存的照片会说明原因并跳过。此前会继续处理,可能导致内存不足。
  • 点击 开始 之后立刻按下的 停止 不会再被忽略。
  • 系统资源紧张时,应用也不会突然退出。
  • 用纯色遮盖带透明部分的图片时不会再透出画面。
  • 保存过程中断时,不会再留下看起来像成品的半截文件。
  • 在 Windows 上,即使路径中含有中文等字符,照片信息同样会保留。
  • 原有的“快速/精确”选择会沿用到新的模型列表。
  • 下载的模型在使用前会经过检查。
  • 已批准的自定义 ONNX 模型文件若发生变化,加载前会再次询问。
  • 处理过程中应用被强制结束或断电时,此前会留下原文件的临时副本。现在会在下次
    启动时清理。

界面语言

  • 视频错误和处理进度的提示现在会以简体中文、韩语和日语显示。此前无论使用哪种
    语言都只显示英文。

模型与许可

  • 内置模型在首次使用时下载,之后即可离线运行。
  • YOLO5Face-n 基于 WIDER FACE 数据训练,应按仅限非商业用途对待。
  • YuNet 采用 MIT 许可,你自己准备的模型遵循各自的条款。

v1.10.2

Choose a tag to compare

@github-actions github-actions released this 22 Jul 06:02

CloakFrame is the new name for Redactly. This release applies the new identity
throughout the app while preserving existing user data.

Improvements

  • Renames the app, executables, packages, namespaces, build targets, assets, and
    translations to CloakFrame
  • Imports settings from earlier Redactly releases on first launch
  • Reuses face and license-plate models already downloaded by earlier releases

v1.10.1

Choose a tag to compare

@github-actions github-actions released this 21 Jul 07:13

Makes very large video processing practical by keeping all temporary data on
the destination drive.

Improvements

  • The private source snapshot and the encoder working file are now stored in
    hidden folders inside the chosen output folder instead of the system
    temporary directory, so processing a large video only needs free space on
    the destination drive
  • Finished videos are published with an instant atomic rename instead of a
    full copy when the temporary location and the destination were on
    different drives, removing tens of gigabytes of extra writes for long 4K
    footage
  • Linux systems with a RAM-backed /tmp no longer run out of space when
    processing videos larger than available memory

All temporary folders are removed automatically when a run completes, fails,
or is cancelled, and the snapshot's tamper protection is unchanged.

Platform notes

  • macOS 15 (Sequoia) is supported again; the macOS 26 requirement introduced
    in v1.10.0 has been lifted

v1.10.0

Choose a tag to compare

@github-actions github-actions released this 21 Jul 06:24

Hardens how Redactly reads and writes files end-to-end and scales processing
up for very large photos and videos.

Security

  • Output files are published atomically inside the chosen output folder using
    symlink-safe, no-replace writes on every platform, so crafted links or files
    swapped mid-run can never redirect or overwrite results outside it
  • Videos are processed from a private snapshot of the source, so edits or
    swaps to the original while a run is in progress cannot affect the output
  • Built-in detection models are verified against a pinned SHA-256 digest
    before they are loaded, and custom ONNX models get stricter shape and type
    validation
  • Image containers (TIFF, PNG, WebP, animated formats) are parsed with strict
    bounds and overflow checks before decoding

Improvements

  • Large media limits are now generous: input files up to 2 GB and images up
    to 512 megapixels are accepted, and the image pipeline adapts its
    parallelism to available memory instead of using a fixed thread count
  • Crowded scenes track more reliably: up to 1024 detections per frame and
    65,536 tracks per video are supported, and license plate detection handles
    denser candidate sets
  • A single unreadable file in a batch no longer slows down processing of the
    remaining items
  • Loaded detection models are reliably reused between runs, including on
    systems where GPU acceleration falls back to the CPU, making repeat runs
    start much faster
  • Detection results with invalid coordinates are filtered out before
    matching, fixing rare inconsistent masking on corrupted frames
  • A face seen clearly in just a single frame is now kept and masked instead
    of being discarded as noise
  • Saving to drives that do not support atomic no-replace renames (exFAT or
    FAT32 USB drives, SD cards, and some network shares) now works on macOS

Platform notes

  • macOS builds now require macOS 26 or newer
  • Windows CI and release builds set up MSVC directly without third-party
    actions

v1.9.1

Choose a tag to compare

@github-actions github-actions released this 14 Jul 04:29

Adds GPU-accelerated detection for Linux source builds and refreshes the
official build stack across all supported platforms.

Improvements

  • Linux source builds can now use CUDA on NVIDIA GPUs or MIGraphX on supported
    AMD GPUs, with automatic CPU fallback when a provider is unavailable or
    rejects a model
  • GPU detection still follows the existing acceleration setting, so it can be
    disabled without rebuilding the application
  • Official builds now use newer, aligned Qt, OpenCV, and ONNX Runtime versions,
    with updated Linux, Windows, and macOS build environments
  • Build requirements are checked explicitly: Qt 6.8.1 or newer and OpenCV
    4.10.0 or newer are required

Linux packaging note

  • The official AppImage continues to use CPU inference. CUDA or MIGraphX
    detection requires a source build linked against a GPU-enabled ONNX Runtime

v1.9.0

Choose a tag to compare

@github-actions github-actions released this 13 Jul 01:18

Adds video review and more anonymization choices, strengthens tracking and
masking, and expands Redactly's language and video-output options.

New features

  • Review detected video tracks on a timeline before encoding and exclude false
    tracks from the entire output
  • Use the new built-in smiley sticker anonymization style
  • Choose H.264 for broad compatibility or HEVC for smaller video files
  • Use Redactly in Japanese; Japanese systems select it automatically on first
    launch, alongside the existing English and Korean options

Improvements

  • Video tracks are matched more safely so masks are less likely to jump to an
    unrelated face
  • Soft mask edges now blend through the padding area while keeping the detected
    region fully covered
  • Korean interface wording has been polished, and the README has been
    consolidated into a single English document

v1.8.1

Choose a tag to compare

@github-actions github-actions released this 10 Jul 04:40

English

Video processing is faster on macOS, and batch runs now protect existing
output files and report incomplete results more clearly.

Improvements

  • Face detection models now use static input dimensions so CoreML can run the
    whole graph instead of leaving unsupported work on the CPU
  • Video output uses Apple's VideoToolbox hardware H.264 encoder when available,
    with automatic fallback to software encoding
  • Redactly checks every planned output before processing and refuses to start
    if a file would be overwritten or two inputs would produce the same path
  • Runs with failures, skipped files, or outputs containing no redacted regions
    now finish as Review required with a clearer summary
  • Very low detection thresholds no longer promote weak video detections into
    thousands of false tracks that obscure unrelated parts of the frame
  • Update notifications now show the release notes in the app's selected
    language, with a choice to update now or postpone

한국어

macOS에서 동영상 처리 속도가 향상되었으며, 이제 일괄 처리 시 기존 출력
파일을 보호하고 완료되지 않은 작업 결과를 더욱 명확하게 알려줍니다.

개선 사항

  • 얼굴 탐지 모델에 고정 입력 크기를 적용하여 CoreML이 일부 작업을 CPU에
    맡기지 않고 전체 그래프를 실행할 수 있도록 개선했습니다
  • 가능한 경우 Apple의 VideoToolbox 하드웨어 H.264 인코더를 사용하며,
    사용할 수 없으면 소프트웨어 인코딩으로 자동 전환됩니다
  • 처리 전에 생성할 모든 출력 경로를 확인하며, 기존 파일을 덮어쓰거나 두
    입력 파일이 같은 경로에 저장될 경우 작업을 시작하지 않습니다
  • 실패하거나 건너뛴 파일 또는 가림 처리된 영역이 없는 출력물이 있으면
    작업 상태를 검토 필요로 표시하고 더욱 명확한 요약을 제공합니다
  • 탐지 임계값이 매우 낮더라도 약한 동영상 탐지가 수천 개의 잘못된 트랙으로
    승격되어 화면의 무관한 영역을 가리지 않도록 개선했습니다
  • 업데이트 알림에서 앱 설정 언어에 맞는 릴리스 노트를 표시하고, 바로
    업데이트하거나 나중으로 미룰 수 있습니다

v1.8.0

Choose a tag to compare

@github-actions github-actions released this 06 Jul 07:35

Video face detection is more accurate: it now analyzes video at the full
detection resolution, so it catches faces it used to miss and stops covering
signage it mistook for a face.

Improvements

  • The fast face model now analyzes video at the full detection resolution
    instead of a third of it, so smaller and partly turned faces that used to
    slip through are covered
  • Fewer false masks: a logo or sign that briefly resembles a face no longer
    stays covered for the rest of the shot
  • Masks follow faces more faithfully — they no longer glide into place a few
    frames early, and stay on a face through brief detection dropouts

Details

  • The bundled fast model ships with a fixed 640 px input; it is now adapted in
    memory at load time to run at the requested video resolution, with a probe
    check and automatic fallback to its native size, or to the CPU, if a GPU
    backend can't run it
  • A track needs several confident detections to be kept, so sparse false
    positives are dropped — while a short but clearly detected face is still
    covered
  • A track that goes too long without a confident detection ends instead of
    coasting indefinitely, and moving faces get more leeway than static signage