Input Manifest

Use a CSV manifest to pass slides to Pipeline or the CLI.

Schema

Column	Required	Notes
sample_id	yes	Unique identifier for the slide; used as the output file stem
image_path	yes	Path to the slide file; absolute paths are recommended
mask_path	no	Path to a binary tissue mask or a multilabel annotation mask; see Annotation-Aware Sampling. For tissue-only sampling, a blank value uses the configured segmentation method
spacing_at_level_0	no	Override for the slide’s native level-0 spacing (µm/px). When blank, slide2vec reads the spacing from the slide file’s metadata
patient_id	no	Required only for patient-level models (see below)

Example

sample_id,image_path,mask_path,spacing_at_level_0
slide-1,/data/slide-1.svs,/data/mask-1.png,0.25
slide-2,/data/slide-2.svs,,

Relative image and mask paths resolve from the working directory, not the manifest’s directory. spacing_at_level_0 may be left blank when the image metadata supplies its physical spacing.

Patient-level manifest

When using a patient-level model (e.g. moozy), add a patient_id column to group slides that belong to the same patient:

sample_id,image_path,patient_id
slide-1a,/data/slide-1a.svs,patient-1
slide-1b,/data/slide-1b.svs,patient-1
slide-2a,/data/slide-2a.svs,patient-2

Slides sharing the same patient_id contribute to one patient embedding. sample_id remains the unique slide identifier. Both identifier columns are read as text, so values such as 0007 retain their leading zeros. Every patient_id must be non-empty after surrounding whitespace is ignored; invalid rows are rejected before tiling begins.

Per-slide embeddings

To also save intermediate slide embeddings under slide_embeddings/, use this CLI configuration:

model:
  save_slide_embeddings: true

The Python equivalent is ExecutionOptions(save_slide_embeddings=True).