SIGMA fp firmware research and on-camera tools. SIGMA fp 韌體研究與機身端工具。
English | 繁體中文 · → ijigen.github.io/fpSup
Everything here runs from an AutoRun.txt on the SD card, in RAM only. Remove the
file, power-cycle, and the camera is stock. Nothing is ever written to flash.
Firmware Ver.5.02 only — a card built for one version writes into whatever
happens to be at those addresses on another.
ijigen.github.io/fpSup — the tools, the releases and the reference, in one page.
fpSup-Merge — pick the
products you want on one card and get AutoRun.txt and fpSup.BIN. Merged cards are
produced here and nowhere else, and every combination is checked against the same
rules the build scripts use. Runs in the browser — no toolchain, no camera.
Two files each — copy AutoRun.txt and the payload container to the root of
the card (fpSup.BIN; releases published before 2026-09-19 name it VSHL.BIN). No
folder to make, nothing to convert, no step afterwards. To put two of them on one
card, use the composer rather than copying both.
| product | version | what it does |
|---|
| fpsup-gyro | v1.14.1test | Writes Gyroflow .gcsv motion and .json lens files while recording CinemaDNG. This test build needs on-camera audio sync validation. |
| fpsup-gyro-base | v1.14.0 | Writes the gyro and accelerometer as a raw .GYR in the root of the disk the take went to — every sample, nothing on the camera but the stream — converted afterwards in a browser. Same code as fpGyroSup v1.14.0. |
| fpsup-gyro2 | v0.1.0 | Writes the fp's gyro and level data into every CinemaDNG frame; the Gyroflow (fpSup) plugin for DaVinci Resolve reads it from the frames and stabilises the clip, with no sync step. Works with Lossless. |
| fpsup-lossless | v0.2.0test | Lossless-compressed CinemaDNG, by the camera's own hardware codec: a Lossless RAW row (SHOOT page 2, CINE) turns it on; frames the codec cannot finish in time are written uncompressed, and compressed clips play back in the camera. Test build. |
| fpsup-og2k | v0.1.5a | The same open gate at 2016×1344, on the sensor's quiet readout at every frame rate. 98 MB/s at 24p 12-bit, 8.3 ms rolling shutter. Test build. |
| fpsup-og3k | v0.2.8a | The sensor's whole 3:2 area at 3024×2010, eight frame rates from 23.976 to 100, native UI in Settings and Quick Set. 221 MB/s at 24p 12-bit — this needs an external SSD, not an SD card. Alpha. |
| fpsup-raw-view | v0.3.0test | RAW monitoring for CinemaDNG 12-bit: a RAW row in the COLOR menu makes the LCD show what will be recorded — recording gain in standby, sensor saturation as white, two latitude curves (SA/GA) mapped onto a 709 screen. The recorded RAW is not changed. Loader v3 card; settings are kept in its own file, not in the camera. Test build. |
| fpsup-res-custom | v0.1.2test | Open Gate formats in one: OG2K, OG3K, OG3.5K, OG4K and S16 as extra movie resolutions in Settings and Quick Set, each a small data file in fpSup\RESCUS; the camera's own tables are not changed. Test. |
| fpsup-screenflip | v0.1.0test | Flip the rear screen 180°, mirror it, or both — set separately for each custom display mode, applied when you switch into that mode. Test build. |
| fpsup-usbshell | v3.4.0test | The shell that answers shl over USB, parasitic on the camera's own PTP gadget so the firmware keeps owning the endpoints. |
The two open-gate builds see the same picture. OG2K is the sensor reduced by three
rather than two, so it is a third of the data and stays on the quiet readout even at 100p.
They cannot share a card — the resolution menu holds three entries and each build takes the
third. Super35/crop must be off for either. Both are pre-release; each release's
README.txt lists exactly what was verified and what was not. Open gate on an fp was first
done by Vitaly Li.
Named fpsup-<product>-v<version>, one directory each, and the tag is spelled
identically — see releases/README.md and
releases/TAGS.md.
The thing most likely to bite: CinemaDNG is limited by write speed long before anything else. Open gate at 24p is about 221 MB/s at 12-bit, 184 at 10-bit and 148 at 8-bit; a mid-range UHS-II card measured here sustains 94, and plain FHD 12-bit already needs 97. A card that cannot keep up buffers in RAM and then stops the take. That is the card, not a bug —
tools/storage-benchmark/measures yours.
Each page says what has been proven, what is being worked on, and what is open.
| # | Project | What it is | Status |
|---|---|---|---|
| 1 | usb shell sup | USB firmware research and data transport | released — v3.2.0. The channel is built on the camera's own PTP gadget, so recording survives it |
| 2 | sensor lab sup | IMX410 modes, ISO, gain, sensor control | prototype — archived r80 overwrites factory data; deployed r81 USER card freezes as REC is pressed. Corrected r83 passes offline checks but awaits card writing and camera retest. ISO/gain explainer |
| 3 | gyro sup | Gyro, six-axis logging, Gyroflow workflow | released — two editions; Base writes a raw .GYR instead, converted in a browser |
| 4 | open gate | Recording the sensor's full 3:2 area | OG3K v0.2.5a alpha · OG2K v0.1.2a test — 3024×2010 or 2016×1344, same field of view; Super35/crop must remain off |
| 5 | 6k to ssd | Getting the best 6K the link can carry | designed; 8-bit lands within 94–100% of native 6K across every remaining unknown |
| 6 | focus sup | DFD, focus model, lens control, follow focus | AF decompiled in depth; no collector built |
| 7 | raw sup | Bayer capture, streaming, compression, packaging | researched; the engine's sustained rate is still the one unmeasured number |
| 8 | ui sup | On-screen display, boot animation | text, colour and a hardware overlay line all work on a camera |
| 9 | power sup | USB-C power delivery and power saving | charging mechanism solved; the rest untouched |
| — | firmware map | Format, subsystems, task ABI, state sources | ongoing — not a product, the ground the rest stands on |
Stopped or paused — kept because what they established still holds
| Project | Why it is here |
|---|---|
| bridge — sigma-fp-bridge | The most complete thing built on the fp's PTP surface. Its measurements of why host-side autofocus is hard, and that UHD 12-bit CinemaDNG cannot come over USB, are load-bearing here |
| gimbal | SIGMA's 0x94xx vendor protocol works during recording, unlike tethered focus. Relative drive only, but absolute position is readable |
| color sup | The colour is reproduced; the tool is awkward and nothing is packaged |
| fpRemote | Not started. Wireless bridge and low-resolution streaming through AutoRun rather than PTP |
- The camera writes its own Gyroflow files while it records. The GCSV streams during the take and the JSON lands a few seconds in; stop is just stop. Zero dropped samples over a 15-minute take, timestamps verified row by row.
- The 12-bit open-gate path works across every ISO, with native Settings and Quick Set UI. 3024×2010 of a 3:2 sensor read, edge to edge — verified by unpacking a real take, not inferred. v0.2.2a adds 8-bit and 10-bit: all three depths were measured recording at OG3K geometry, where 8 and 10-bit previously fell back to UHD30 without any on-screen sign. v0.2.1a's four-callsite shutter-angle fix and loader instruction-cache synchronization carry forward. v0.2.3a fixes v0.2.2a's format-table pass-through pointer; all 100 DNGs in exact no-shell OG3K/UHD/FHD 25p/180° tests recorded at 1/50 with correct geometry. Super35/ crop must remain off. 8/10-bit playback and highlight headroom, other frame rates, sustained-media and playback-with-UI regressions remain.
- The IMX410 ISO and gain chain is fully decompiled, with an explainer that separates firmware-confirmed behaviour from OTP values still needing measurement.
- The USB shell does not break recording. It is built on the camera's own PTP gadget, so the firmware owns the endpoints and re-creates them after a record-mode reconfiguration.
docs/SHELL_COMMANDS.md |
the firmware shell's 77 commands, with usage text asked from a live camera |
docs/SHELL_CAPABILITIES.md |
what those commands reach — memory and I²C writes, menu setters, sensor readout modes |
docs/MENU_MODIFICATION.md |
every menu item carries its own metadata in ROM, right after its name string |
docs/FREEZE_ROOTCAUSE.md |
why the first USB shell froze the camera, mechanism and all |
BUILDING.md |
how a card works and how to build, combine, test and release one — the walkthrough (中文) |
SUP_BUILD_RULES.en.md |
the build contract every sup follows (中文) |
HOOKS_AT_POWER_OFF.md |
why a hook must come out at power-off, and how |
ijigen.github.io/fpSup —— 工具、釋出版、參考資料都在一頁。
fpSup-Merge —— 勾選要的產品,
產生 AutoRun.txt 與 fpSup.BIN。合併版只在這裡產生,而且每個組合都用建置腳本
同一套規則檢查過。在瀏覽器裡跑 —— 不用工具鏈,不用相機。
每份就兩個檔 —— AutoRun.txt 與 fpSup.BIN(2026-09-19 之前發布的版本叫 VSHL.BIN)放進卡片根目錄。不用建資料夾、不用轉檔、事後沒有步驟。要把兩份放同一張卡,用合併器,不要兩份都複製。
| 產品 | 版本 | 做什麼 |
|---|
| fpsup-gyro | v1.14.1test | 錄製 CinemaDNG 時寫出 Gyroflow 的 .gcsv 運動紀錄與 .json 鏡頭檔。此測試版的聲畫同步仍待實機驗證。 |
| fpsup-gyro-base | v1.14.0 | 在錄影那顆磁碟的根目錄寫一個原始 .GYR(陀螺儀與加速度計每個樣本都在),機上只做串流,事後在瀏覽器轉換。與 fpGyroSup v1.14.0 同一份程式。 |
| fpsup-gyro2 | v0.1.0 | 把 fp 的陀螺儀與水平儀資料寫進每一格 CinemaDNG;DaVinci Resolve 的 Gyroflow (fpSup) 外掛直接從畫面檔讀取並防震，不用手動同步。可與 Lossless 同卡。 |
| fpsup-lossless | v0.2.0test | 用相機自己的硬體編碼器錄無損壓縮 CinemaDNG:SHOOT 第 2 頁(CINE)多一列 Lossless RAW 開關;來不及壓的畫格照原樣寫入未壓縮,壓縮過的片段可在機內回放。測試版。 |
| fpsup-og2k | v0.1.5a | 同樣的 open gate,2016×1344,每個幀率都走感光元件的安靜讀出。24p 12-bit 為 98 MB/s,捲簾 8.3 ms。測試版。 |
| fpsup-og3k | v0.2.8a | 感光元件完整的 3:2 面積,3024×2010,八個幀率從 23.976 到 100,Settings 與 QS 有原生 UI。24p 12-bit 為 221 MB/s ——這個碼率要外接 SSD,SD 卡不夠。Alpha。 |
| fpsup-raw-view | v0.3.0test | CinemaDNG 12-bit 的 RAW 監看:COLOR 選單多一列 RAW,螢幕顯示即將錄下的內容——待機用錄影增益、感光元件飽和即白、兩條寬容度曲線(SA/GA)映射到 709 螢幕。錄下的 RAW 不被改動。Loader v3 卡;設定存在自己的檔案,不寫相機設定。測試版。 |
| fpsup-res-custom | v0.1.2test | Open Gate 各格式合一:OG2K、OG3K、OG3.5K、OG4K、S16 成為 Settings 與 QS 的錄影解析度新選項,每個格式是 fpSup\RESCUS 裡的一個小資料檔;不改相機原廠的任何表。測試版。 |
| fpsup-screenflip | v0.1.0test | 背螢幕翻轉 180°、鏡像或兩者 —— 每個自訂顯示模式各自設定,切換進那個模式時才套用。測試版。 |
| fpsup-usbshell | v3.4.0test | 透過 USB 回應 shl 的 shell。寄生在相機自己的 PTP gadget 上,端點仍由韌體管。 |
兩個 open gate 看到的是同一個畫面。 OG2K 是把感光元件縮三倍而不是兩倍,所以資料量只有三分之一,
而且連 100p 都還在安靜讀出上。兩者不能放同一張卡 —— 解析度選單只有三格,各自佔用第三格。
**兩者都必須關閉 Super35/crop。**都還是預覽版:各自的 README.txt 寫明了驗過什麼、沒驗什麼。
最早在 fp 實現 open gate 的是 Vitaly Li。
命名是 fpsup-<產品>-v<版本>,一個版本一個資料夾,tag 逐字相同 ——
見 releases/README.md 與 releases/TAGS.md。
最可能咬人的一件事: CinemaDNG 遠在其他東西之前就先被寫入速度卡住。 Open gate 24p 在 12/10/8-bit 各約 221 / 184 / 148 MB/s;這裡實測一張中階 UHS-II 卡持續寫入 94,而純 FHD 12bit 就已經要 97。撐不住的卡會先用 RAM 緩衝,然後停止錄影 —— 那是卡不是 bug。用
tools/storage-benchmark/量自己的。
每一頁都寫清楚已經證實了什麼、正在做什麼、還有什麼沒解。
| # | 項目 | 是什麼 | 狀態 |
|---|---|---|---|
| 1 | usb shell sup | USB 韌體研究與資料傳輸 | 已釋出 —— v3.2.0。通道建在相機自己的 PTP gadget 上,所以錄影撐得過去 |
| 2 | sensor lab sup | IMX410 模式、ISO、增益、感光元件控制 | 原型 —— 封存 r80 覆寫原廠資料；已寫卡的 r81 USER 版按 REC 立即凍結。修正後 r83 通過離線檢查，尚待寫卡與相機複驗；ISO／增益說明 |
| 3 | gyro sup | 陀螺儀、六軸記錄、Gyroflow 流程 | 已釋出 —— 兩個版本;Base 版寫原始 .GYR,在瀏覽器裡轉換 |
| 4 | open gate | 錄下感光元件完整的 3:2 面積 | OG3K v0.2.5a alpha · OG2K v0.1.2a test —— 3024×2010 或 2016×1344,視角相同;Super35/crop 必須關閉 |
| 5 | 6k to ssd | 把鏈路載得動的最好 6K 拿出來 | 設計完成;8bit 在所有剩餘未知數下都落在原生 6K 的 94–100% |
| 6 | focus sup | DFD、對焦模型、鏡頭控制、跟焦 | AF 深度反編譯完成;收集器還沒做 |
| 7 | raw sup | Bayer 擷取、串流、壓縮、封裝 | 研究過;引擎的持續速率仍是唯一沒量到的數字 |
| 8 | ui sup | 螢幕顯示、開機動畫 | 螢幕出字可行;顏色編碼未解 |
| 9 | power sup | USB-C 供電與省電 | 充電機制已解;其餘未動 |
| — | firmware map | 格式、子系統、任務 ABI、狀態來源 | 進行中 —— 不是產品,是其他東西站著的地面 |
- 相機自己在錄影當下寫出 Gyroflow 要的檔案。 GCSV 邊錄邊串流,JSON 開始幾秒後落地,停止就只是停止。15 分鐘的 take 零掉樣,時間戳逐列驗過。
- Open gate 的 12-bit 路徑已驗證全 ISO,原生 UI 已整合。 3:2 讀出的 3024×2010,整幅邊到邊 —— 解檔實錄驗證,不是推論。v0.2.2a 新增 8-bit 與 10-bit;v0.2.3a 修正 format-table pass-through 指標。精確 no-shell 的 OG3K/UHD/FHD 25p/180°三段共100張 DNG 全為1/50且幾何正確。OG3K 必須關閉 Super35/crop。8/10-bit 的回放與高光餘裕、其他幀率、長時間寫入與新 UI runtime 回放仍待回歸。
- IMX410 的 ISO 與增益鏈完整反編譯,說明頁把「韌體確認的行為」跟「還需要量測的 OTP 值」分開。
- USB shell 不會弄壞錄影。 它建在相機自己的 PTP gadget 上,端點由韌體管,錄影模式重配之後韌體會自己重建。
BUILDING.zh.md |
卡片怎麼運作、怎麼建卡、合併、測試與發布 —— 完整流程說明(English) |
SUP_BUILD_RULES.md |
每個 sup 都要遵守的建置規範(English) |
HOOKS_AT_POWER_OFF.md |
為什麼 hook 關機時必須拆掉、怎麼拆(英文,附中文摘要) |