Batch Processing Tab

The Batch tab runs reconstructions on many datasets in parallel, with multi-GPU execution, AI Reco integration, COR outlier / missing-value correction, and remote-host support.

Features

• file list with per-row checkbox, COR input, status, and series tint

• Shift-click range selection

• Batch Try, Batch Full, Batch AI Reco with per-phase selection checkboxes: Try / Infer / Full / TomoLog

• Fix COR Outliers with series grouping, configurable max delta, and automatic series-mean fill for missing CORs

• Delete Selected with confirmation

• per-file streaming status during inference; live COR updates

• unified GPU queue: one reconstruction / inference process per GPU slot

• remote SSH execution

File table

Col	Name	Purpose
0	☑	Select / deselect (Shift-click for a range)
1	File	Filename (relative to the data folder)
2	COR	Per-row COR input
3	Status	Queued / Running / Inferring / Uploading / Done / Failed

Rows are tinted by their filename series.

Right-click on a row opens a context menu with Edit parameters, Apply parameters to selected, View Try / View Full, and Delete from list.

Selection

• Click a checkbox — toggle a single row

• Shift-click — toggle every row between the previously clicked row and this one (no automatic COR filling — use Fix COR Outliers for that)

• Header checkbox — select / clear all rows

Batch Try / Full

Batch Try and Batch Full drop every checked row into the GPU queue. One TomoCuPy subprocess per GPU slot, pinned via CUDA_VISIBLE_DEVICES. The top-bar Try COR input is the fallback for any row whose own COR is empty; if both are missing for a row, the run is blocked with a clear error.

Batch AI Reco

All phases use the same shared GPU queue — one file per GPU at a time, next one dispatched as soon as a slot frees up. Four checkboxes next to the button let you pick which phases to run:

• Try (default on) — Phase A

• Infer (default on) — Phase B

• Full (default on) — Phase C

• TomoLog (default off) — Phase D

Skipping Try assumes try_center TIFFs already exist on disk. Skipping Infer makes Full use whatever CORs are already in the table.

Phase A — Try

Try reconstruction for every checked row.

Phase B — Inference

One python -m tomogui._infer_worker per file, per GPU slot. Worker prints [infer-worker] OK <path> => <cor>; the GUI updates that row’s COR and status the moment it sees the line. A hung file only blocks its own GPU slot — the rest of the queue keeps flowing.

Phase C — Full

Full reconstruction for every row that now has a COR.

Phase D — TomoLog upload (optional)

Tick the TomoLog phase checkbox. After Phase C, every file whose Full succeeded is uploaded to TomoLog using the settings in the right-hand TomoLog panel. Uploads are sequential.

Fix COR Outliers (also fills missing CORs)

Two passes run on one click, both using the same filename-series grouping (^(.*?)[._-]*(\d+)$):

1. Outlier replacement — in each series, flag any COR outside median ± min(max_delta, max(10, 5·MAD)) and replace it with the average of the two nearest non-flagged neighbours by index.

2. Missing-COR fill — any selected row still empty is filled with the mean of existing CORs in its series across the whole table (donors can be checked or unchecked, anywhere in the list). 0 donors → left empty and reported.

max_delta is the Max COR delta spinbox (default 50 px).

Auto-skip undersized files

Within a series, any file with noticeably smaller HDF5 /exchange/data than its peers is marked skipped automatically — typically an aborted acquisition.

Delete Selected

Removes checked rows from the file list and table after confirmation. Does not touch anything on disk.

Remote execution

Set Remote host in Advanced Configuration. The batch queue SSHes to the remote host, runs TomoCuPy / _infer_worker there with CUDA_VISIBLE_DEVICES per worker, and streams stdout back.

See SSH Setup for Remote Machines and GPU Management.