Radar

Build a model that labels every cell of a radar range-azimuth map as background, suitcase, chair, human or wall from six radar heatmaps.

Semantic segmentation

The task

Radar systems send out signals and read their reflections to locate objects, but noise and echoes from surrounding clutter make this hard. This task uses data from an indoor radar experiment and asks for semantic segmentation of what the radar sees.

Each sample is a 7 × 50 × 181 tensor: six normalised heatmaps (static and dynamic versions of the range-azimuth, range-elevation and range-velocity maps) plus a label map in range-azimuth form. Labels are -1 for background, 0 for suitcase, 1 for chair, 2 for human and 3 for wall.

From the six heatmaps you must predict the 50 × 181 label map. Scoring rewards correctly labelled object cells far more than correctly labelled background. This on-site task extends the at-home Radar problem, which only asked whether a human was present.

Abridged by SOTA from the official materials. The official statement has the exact rules, and it wins wherever this summary differs.

At a glance

You get

Samples stored as .mat.pt tensors of shape 7×50×181 (six heatmaps plus the label map); 1,800 training, 500 validation and 500 test samples.

You submit

A 50×181 label map per sample. The submitted notebook must produce submission.zip containing submission_val.csv and submission_test.csv, each with a header, a filename column and columns pixel_0 to pixel_9049.

Scoring

Weighted pixel accuracy: each correctly identified background pixel earns 1 point and each correctly identified non-background pixel earns 50 points; the total is divided by the maximum possible score (0 to 1). The Scientific Committee's highest score (0.90) and baseline score (0.67) on Leaderboard B were used for score normalisation.

Rules

Notebook must finish within 15 minutes on the test machine
At most 15 successful submissions; up to 2 can be selected and the higher Leaderboard B score counts
Only a notebook may be submitted (no self-made mounted datasets or .pth files)
Test machine has no internet access
Pretrained models may be used only if they import without a network connection

Format

Individual Contest, Day 1 (5 August 2025); six hours for the day's three tasks

Links

Radar

The task

At a glance

Links

Notebook

Data

Solution

Results

Source files

Details