README.md

---
license: cc-by-4.0
datasets:
- risenyard/egms-qa-dataset
tags:
- safetensors
- remote-sensing
- insar
- ground-motion
- time-series
- egms
---

# EGMS-QA Encoder

Pretrained encoder for EGMS displacement time series within 7 km tiles. It
produces 256-dimensional point representations, which are spatially pooled
into 65 tile tokens.

This repository contains the released weights, model settings, training recipe,
and evaluation results.

## Use the model

See the [architecture on GitHub](https://github.com/risenyard/egms-qa/blob/main/src/egms_encoder/README.md#architecture),
then follow [installation and a one-tile run](https://github.com/risenyard/egms-qa/blob/main/src/egms_encoder/README.md#installation) to try the encoder.
The same guide covers [local inputs](https://github.com/risenyard/egms-qa/blob/main/src/egms_encoder/README.md#use-local-inputs)
and [training, resuming, and token extraction](https://github.com/risenyard/egms-qa/blob/main/src/egms_encoder/README.md#reproduce-training).
Prepared tiles and precomputed tokens are available in the
[Dataset repository](https://huggingface.co/datasets/risenyard/egms-qa-dataset#encoder-data).

## Files

| file | purpose |
|---|---|
| [encoder.safetensors](encoder.safetensors) | encoder weights |
| [config.json](config.json) | model architecture and input dimensions |
| [normalization.json](normalization.json) | input mean, standard deviation, and residual scale |
| [training_args.json](training_args.json) | training recipe and checkpoint-selection record |
| [eval_results.json](eval_results.json) | masked-reconstruction metrics |

Inference requires the weights, model config, and normalization. The Dataset
repository provides the measurements and split manifest.

## Input requirements

| input or output | contract |
|---|---|
| tile displacement | vertical displacement in mm, `[N,294]` |
| coordinates | EPSG:3035 easting and northing in meters, `[N,2]` |
| model preprocessing | checkpoint normalization and centered coordinates |
| point representations | `[N,256]` |
| pooled tokens | `[65,256]` and a 65-element validity mask |

The Dataset stores `[0,294)`, corresponding to `[8,302)` on the 304-step
source-preparation axis. Its data config retains the source offset and six-day
cadence for physical-time calculations.

Use this checkpoint's normalization for inference. When training a new encoder
on another corpus, fit normalization on that corpus's training split.

## Evaluation

Evaluation covers 1,000 held-out tiles with 2,047,451 point histories. A central
88-step interval, approximately 30% of the 294-step input, is masked at the
same positions for every point in a tile.

| metric | value |
|---|---:|
| normalized MSE | 0.0702 |
| MSE | 2.279 mm² |
| RMSE | 1.510 mm |
| MAE | 1.007 mm |
| pooled EGMS residual | 1.433 mm |
| per-point RMSE P10 / P50 / P90 | 0.54 / 1.03 / 2.43 mm |

These values describe reconstruction of held-out observations under the
specified masking protocol.

## Scope and license

The encoder consumes prepared EGMS-QA tiles. Official-product downloading,
format conversion, and preparation of another reference period require a
separate workflow. Its outputs describe observed deformation and do not
establish causes, predict future motion, or certify structural safety.

The encoder is released under CC-BY-4.0.