cellmap_flow.portable.volume_calls

What the browser calls to plan an annotation volume and its good regions, and to tell whether a model trains on the session’s volume as it is: the functions web/src/finetune/volume.ts, web/src/finetune/good-regions.ts, web/src/finetune/run.ts and web/src/backend/routes/finetune.ts’s submit run through chunkmirage’s Engine.call("cellmap_flow.portable.volume_calls:<name>", args, arrays) (calls.py says the shape). Each takes plain numbers in its JSON and answers JSON; the rules are volume.py’s and regions.py’s, which the server calls in its process. A refusal (a ValueError of those rules) is answered as {"error": <its sentence>}, so the page shows the sentence the server’s route would, not a Python traceback.

Functions

plan_volume(args, arrays)	volume.plan_volume of level (shape, voxel, corner:
plan_paint_volume(args, arrays)	volume.plan_paint_volume of level (shape, voxel,
shown_level(args, arrays)	{"index": volume.shown_level(voxel_sizes, pixel_nm)}: which of
training_refusal(args, arrays)	{"error": <the sentence>} when volume (its record) is not
carry_note(args, arrays)	Training's submit with a model whose grid is not the session
mark_region(args, arrays)	The region a mark makes (regions.new_region): a box of the size
region_boxes(args, arrays)	The good_regions layer's boxes of regions: per entry its
anchor_boxes(args, arrays)	regions.anchor_boxes of regions on the grid of volume

Module Contents

cellmap_flow.portable.volume_calls.plan_volume(args, arrays)

volume.plan_volume of level (shape, voxel, corner: the raw level the model reads) and model (its voxel sizes and its blocks, as block_shapes takes them), with dtype, the labels’ (volume.annotation_dtype of output_type, what the model trains on).

cellmap_flow.portable.volume_calls.plan_paint_volume(args, arrays)

volume.plan_paint_volume of level (shape, voxel, corner: the raw level read at the voxel size) and voxel_size (nm, z, y, x), a volume painted before any model, with dtype uint8: its labels are background and objects, which a model’s volume they are carried to holds in its own dtype.

cellmap_flow.portable.volume_calls.shown_level(args, arrays)

{"index": volume.shown_level(voxel_sizes, pixel_nm)}: which of the raw’s levels (voxel_sizes, nm, finest first) the viewer shows at a screen pixel of pixel_nm (null: not known, the finest), the default voxel size of a volume painted before any model.

cellmap_flow.portable.volume_calls.training_refusal(args, arrays)

{"error": <the sentence>} when volume (its record) is not trained on (volume.training_refusal: a paint-first volume), else {}.

cellmap_flow.portable.volume_calls.carry_note(args, arrays)

Training’s submit with a model whose grid is not the session volume’s: {"note": volume.carried_note(...)} when to (the volume planned for the model to train) is on another grid than from (the session’s volume, its paint’s), each a volume as the browser keeps it, and kept where the old one is kept; {} when they are on one grid (volume.same_grid), and the model trains on the volume as it is.

cellmap_flow.portable.volume_calls.mark_region(args, arrays)

The region a mark makes (regions.new_region): a box of the size size_nm asks for (regions.region_size_nm, the volume’s output block by default: volume has its chunk_size and output_voxel_size, or is null) centred on centre_nm, labelled label or after the regions already stored. Answers the region, and boxes, the layer’s boxes of the stored regions and it (region_boxes).

cellmap_flow.portable.volume_calls.region_boxes(args, arrays)

The good_regions layer’s boxes of regions: per entry its offset_nm, shape_nm and label, or null for one that is not a region (regions.region_box_nm), so the layer’s ids stay the entries’ indices.

cellmap_flow.portable.volume_calls.anchor_boxes(args, arrays)

regions.anchor_boxes of regions on the grid of volume (its dataset_offset_nm, output_voxel_size, dataset_shape_voxels and chunk_size, the model’s output block): the boxes, and the entries skipped, each by its index in regions.