v3.4.0

📝 Documentation

• build with Zensical, deploy with GitHub Pages actions
• orange header on Zensical's modern theme

🔄 Other

• 📝 Simplify the documentation
• ✨ Store the whole processing config with each label image
• 🐛 Stop per-tile decisions from cutting cells at tile seams
• 🐛 Fold 3-D Cellpose fragments back into their cells
• 📝 Show the Cellpose normalize and fragment settings in the Cellpose configs
• ✨ Record Cellpose's filled-in defaults and the tools' versions
• ✨ Redo a segmentation by deleting its labels from the store
• 🐛 Install the workflow's patchworks from the clone itself
• ✨ Zip the finished store by default in multi runs
• 🐛 Run the bundle job with the relate jobs' QOS and partition
• ♻️ Shorter configs and comments: one table of defaults
• 🐛 PlantSeg: standardize every tile with the image's statistics
• 🐛 IoU stitching: match a cut-off view to the whole object
• 🐛 Judge fragments against the diameter, the same in every tile
• 🔊 Log how well the tiles agree across the seams (IoU stitching)
• ✨ Launcher: pixi workspace, cluster file browser, clearer layout, plan all
• 🐛 Launcher: right pixi environment, convert only, no stale config warning
• 🐛 Viewer: relations on hover, zip bundles, no axes guess
• 🐛 Viewer workspace: use the clone's own patchworks
• 📝 List every setting in the configs, commented with its default
• ✨ Launcher: pick a folder of single-plane TIFFs as the input
• Update scicore URL
• ✨ Launcher: stitch per segmentation (touch for nuclei, iou for cells)
• 🥅 Check every config before converting, dry runs included
• ✨ pixi run status: how far a run has got
• 🔧 Own configs in config/local/, launcher files ignored, SSH keepalive
• ✨ pixi run check: a config's problems and the run's plan, before running
• ✨ Launcher: check on the cluster, resources per run, a Jobs tab that acts
• ✨ Launcher: reopen a past run's config in the form
• 🔧 Profile: one memory value for the segment and prepare jobs
• 📝 Launcher docs: past runs, resources, check and plan, the Jobs tab
• ✨ Launcher: partitions, QOS, accounts and GPUs as dropdowns from the cluster
• ✨ Launcher: PlantSeg and the watershed as methods, cells grown from nuclei
• ✨ Fetch every model before the run, on the login node
• ⚡ Size Cellpose's "auto" tiles for the GPU the segment jobs get
• ✨ Try the method on real tiles: a picture of one, its objects outlined
• ✨ See the channels: a picture of each, to pick them by
• ✨ Say why a step failed and what to do, and why jobs wait
• ✨ Launcher: the Cellpose model from what the cluster's Cellpose has
• ✨ Launcher: save the settings as a named template, start the next run from it
• ✏️ Launcher: gpu_memory_gb 0 means the segment jobs' GPU now
• 🐛 Review: merging objects did nothing under pandas 3
• 💚 Test: the launcher's work_dir check gets a cluster path on Windows too
• ✨ Launcher: every setting explains itself, with an example
• 🐛 Launcher: say when the cluster's workflow is older than the launcher
• ✅ Test: the fake clone's files are live, as a reused SFTP channel sees them
• check a blank QOS against the account's default QOS
• one-click fix for a controller QOS that is too short
• Mail only the last attempt's failure, and only the first start
• Jobs tab: live step progress, step logs, convert-only view
• Jobs tab: converted only once finished; smoother live refresh
• Jobs tab: why earlier attempts failed; live tiles, merge and relate
• write huge source chunks slab by slab to bound memory
• build every pyramid level from one read of each slab
• Mail what a finished run made: the conversion alone, or the labels
• the controller job in one plain line, its settings tucked away
• name a run's SLURM jobs after what it does
• Jobs tab: queue names without the executor's run id
• Jobs tab: see checkpoint steps; stay within the server's SSH channels
• download the weights, do not load the model
• a path needs no cellpose check; let _get_model do it
• annotate the one-pass pyramid's level lists (mypy)
• a dropped SSH connection asks to log in again
• Jobs tab: send the log paths on stdin, not in the command
• remember the connection; log in with an SSH key
• give cudaDecon its integer settings as integers
• Jobs tab: cancel all of a run's jobs
• Jobs tab: a Cancel jobs panel that is always there; progress first
• Jobs tab: the failures in one folded panel
• Fetch Cellpose weights on 3, 4.0-4.1 and 4.2; clear jobs and failures
• Jobs tab: say why a run stopped; this run's failures; no false DONE
• take over a killed controller job's marker
• say in its terminal why the SSH connection dropped
• remove the stage store last; finish a merge stopped after it
• colorama on Windows
• show a launch happening; say what a seeded segmentation waits for
• start a seeded segmentation once its seeds' labels are ready
• auto" sizes it from the objects themselves
• Jobs tab: say when a try waits for a GPU
• A tile try says what it does as it goes
• A tile try says its tile size, overlap and device
• A tile try gets the segment jobs' time, and stops in time to report
• Tile try: the fullest tiles, the busiest plane, a readable page
• See the channels: zoom into a region, with full detail
• keep each segmentation's Snakemake output and how it ended
• unbuffered Snakemake output, and an order-proof tee test
• import patchworks from the clone's src/, not the installed copy
• say how much memory ran out and what to enter next time
• Multi runs: don't check configs again in each Snakemake; 16G controller
• tie a failed step to its own segmentation; portable signal names
• mirror each tile out before deconvolution
• Object table: measure blocks in processes, aligned to the image's chunks
• faster stitching, level 0 sharded as written, streamed label pyramid
• Segment and convert: overlap reading and writing with the work
• CPUs per step in Cluster resources
• ask scontrol when squeue cannot say if the driver lives
• 🐛 Status: object count from the label group
• 🐛 overlap "auto": size the halo from the Cellpose diameter
• Logo as SVG, used by the README and the docs
• smooth thread strokes, true spool rectangle and screw circles
• 🐛 DoG: threshold as a fraction of the image's intensity range
• 🐛 launcher: a slow model list no longer crashes the form
• 🐛 run_multi: ask the other login node whether its driver still runs
• 🐛 run_multi: record the launcher controller's job; read pam_slurm_adopt
• ✨ run_multi: release a killed run's Snakemake locks at start
• ✨ launcher: a reload (F5) keeps the SSH connection
• 🐛 bundle: drop the out-of-date archive first when both do not fit
• 🐛 launcher: one click on Launch is enough
• 🐛 launcher: a partition's own QOS (AllowQos)
• 🐛 status: a running Cellpose job is not "failed"
• 🐛 object table: size the workers for the image chunks they decompress
• 🐛 IoU stitching: a sliver under the size filter rejoins its cell
• ✨ launcher: the relations and the bundle in Progress
• ✨ launcher: relations and the bundle in Steps; no step stuck "submitted"
• ✨ viewer: pixi run bench, where napari's time goes
• ⚡ viewer: a layer per channel, loading in the background
• 🐛 bench: time the viewer's reads, not reopening the zip; --render works
• ⚡ viewer: read zarr directly, open a bundle once
• 🐛 viewer: channels back on one layer's slider, no background loading
• ⚡ viewer: stored label colours; opening says what takes how long
• ⚡ viewer: object tables after the window opens; say when the env is remote
• 🐛 viewer: back to dask for the image and labels
• 🐛 viewer: a zip reader per level again; --no-hover
• ✨ viewer: a layer per channel, only the first shown; named sliders
• ✨ channel names: kept at conversion, shown as napari's layer names
• ✨ channel names from a folder of single-plane TIFFs
• ✨ relations: progress per relation, and in the launcher's bar
• ✨ patchworks sizes / filter: pick and apply a min volume after a run
• ✨ sizes / filter: by length, and say when volume has no gap
• 🐛 CI: skip the slider test without napari; a type mypy rejected

👥 Contributors

• @claude
• @lguerard