Embedding map
Lays the embeddings out on a plane (UMAP or PCA): clusters, classes that mix and odd images, with the flags of the checks.
How it works
Images that look alike sit together. UMAP keeps neighbours together best (used when installed, or chosen); PCA is exact and fast. Each point carries its label and the flags of the tables connected (quality, duplicates, labels), for a Chart (points, x and y) or a Table.
Inputs
| Input | Takes | What it is |
|---|---|---|
Embeddingsembeddings | Files npy, json | embeddings.npy and paths.json of Embed images |
Flagsflags | Table csv | Tables of the checks, to mark the flagged images Optional. Takes several connections. |
Outputs
| Output | Gives | What it is |
|---|---|---|
Mapmap | Table csv, json | embedding_map.csv: path, x, y, label, flags |
Live telemetrytelemetry | Live telemetry | What the block sends while it runs, for views such as Chart or 3D view. |
Settings
| Setting | Default | What it does |
|---|---|---|
Methodmethod |
auto | One of: auto, UMAP, PCA |
Neighbours (UMAP)neighbours |
15 | From 2 to 200 |
Seedseed |
1 |
Where it runs
On one of your computers with the EsaSim worker: Any worker.
The worker installs the Python packages it needs when they are missing: numpy.
In its code, settings are ctx.params["key"], inputs ctx.inputs["id"] and outputs ctx.output("id", "name.ext"); see Custom blocks.
More in Processing
Curate and splitRemoves the images the checks flagged (or a person rejected), balances the classes and splits train and validation, with a before/after report.Find duplicatesFinds copies of the same picture (resized, recompressed, slightly cropped) and, with embeddings, images that show the same thing; keeps one per group.Image previewsSmall previews of the images to look at (the flagged ones, a sample of the rest), with what the tables say about each, for a Review grid.Image quality checkOpens every image and measures its size, sharpness, exposure and contrast; flags broken, small, blurred, dark, bright or flat images with the reason.Labels to 3D modelsTurns labelled volumes (a CT segmentation, any label map or masks) into a clean 3D model per structure: one coloured GLB scene per case for the 3D view, and STL files for 3D printing.Python scriptYour own Python code: read inputs, compute, write outputs.