Recent HistoPilot upgrades

I have been organizing the HistoPilot stack recently, and I am also considering whether the project name on the homepage should be unified as HistoPilot. A single public entry point would make the relationship between PathTogether, the plugin, and HistoPilot easier to understand. This is only a branding and entry-point consideration for now; it does not mean renaming the internal repositories, services, and compatibility contracts all at once.

Multicolor fluorescence compatibility

This upgrade also adds a clearer compatibility direction for multicolor fluorescence images. The important part is not to treat every image as ordinary RGB, but to preserve channel order, channel names, display colors, intensity windows, gamma, T/Z selection, and coordinate information. The channel state shown by the viewer must match the image semantics actually sent to HistoPilot.

Display images and AI derivatives need separate encoding identities. Multichannel fluorescence images and thumbnails should prefer 4:4:4 to protect color boundaries; caches and checkpoints must also distinguish render context from encoder version. Changing pseudocolors or display settings must not make the AI silently believe it saw a different image. Marker names should only be recorded when the source metadata or validation supports them.

COS and Baidu Netdisk for bandwidth routing

For bandwidth, I am considering two optional ingestion paths: browser-side multipart upload to COS, followed by a worker that ingests the object locally, and a controlled Baidu-share flow covering parsing, enumeration, user selection, download, validation, and import. Large files should not repeatedly pass through the application API, and HistoPilot or a web request thread should not own a long-running download.

These are still compatibility and routing proposals that need real validation for throughput, resume behavior, cost, quotas, cleanup, and retries. Cloud storage or a premium netdisk plan cannot remove the physical bottleneck on the server's final network segment, so the data path and safety boundaries come first.

In short: make the homepage feel more like a HistoPilot entry point, treat multicolor fluorescence as real structured image data, and move large-file traffic onto recoverable storage paths where appropriate.

The author was lazy; this note was published by GPT.