◈ PCA 3D: Training Points

🔬 Classify Embeddings

Point sampling • Cosine similarity classifier • Viewport-adaptive Select a class then click the map to add training points
CORS / network error. The S3 bucket needs:
Access-Control-Allow-Origin: *
Access-Control-Allow-Headers: Range, Content-Type
🗺 S2 Tile Grid
Loading tile grid…
Step 1 · Load Embedding Tile(s)
STAC Catalog: embeddings
Click Fetch to load items
Embedding Layers (128-band) 0
Load a tile from STAC above
Step 3 · Classify Viewport
🔬 Supervised
75%
Sample first, then classify
🔀 Unsupervised (K-means)
auto-set from classes
75%
Select tile and run K-means
Step 2A · Add Class Samples
?
Expected CSV format:
class_name,latitude,longitude[,color]

• Header row auto-detected
• Separator: comma, tab, or semicolon
• Recognized names: class, label, lat, latitude, lon, longitude, color
• Optional color column: hex like #e63946 — restores class colors on re-import

Example:
Wheat,51.234,-1.456,#f4d03f
Forest,52.100,-0.800,#27ae60
Classes 0
No classes yet
Select a class above to add points
Step 2B · Sample Training Points
?
Sampled Points CSV (with embeddings):
class_name,latitude,longitude,color,emb_1,emb_2,…,emb_128

• Export saves every sampled point with its full embedding vector
• Re-importing restores classes, colors, map markers and embedding values — no need to re-run sampling
• Add new points on the map, then use Run Sampling (new only) to append their vectors before exporting again
Select tiles to sample from (point must fall within tile extent):
Load embedding tiles first
Add class points and load a tile to sample