AMAP-TECH Algorithm Competition: Dynamic Road‑Condition Analysis from In‑Vehicle Video Images

Alibaba Amap (AMAP‑TECH) launches an algorithm competition on July 8. The challenge is to develop AI‑driven computer‑vision methods that infer real‑time road‑condition status (smooth, slow, congested) from short sequences of in‑vehicle video frames.

Background

Gaode Map provides massive positioning and navigation services. Traditional traffic‑state estimation relies on vehicle trajectories, which become unreliable on low‑traffic or abnormal driving patterns. Video frames captured by dashcams contain richer visual cues (vehicle count, road width, occupancy) that can improve traffic‑state accuracy.

Problem Definition

Given a sequence of 3‑5 images with GPS timestamps, where one frame is designated as the reference frame, participants must output the traffic‑state label (smooth, slow, congested) for the entire sequence, using the reference frame as the ground‑truth reference.

Data Description

The dataset is split into a preliminary (pre‑competition) set and a final (post‑preliminary) set. The preliminary set contains 1 500 sequences (≈ 7 000 images) for training and 600 sequences (≈ 2 800 images) for testing. Road‑state distribution: smooth 70 %, slow 10 %, congested 20 %. Labels are primarily derived from the reference frame, with additional context from surrounding frames when vehicle density is high.

Data Format

Each sequence is stored in a folder containing the reference frame and its neighboring frames (max 5 frames). Annotation files are provided in JSON, e.g.:

{
    "annotations": [
        {
            "id": "000001",
            "key_frame": "2.jpg",
            "status": 0,
            "frames": [
                {"frame_name": "1.jpg", "gps_time": 1552806921},
                {"frame_name": "2.jpg", "gps_time": 1552806926},
                {"frame_name": "3.jpg", "gps_time": 1552806931},
                {"frame_name": "4.jpg", "gps_time": 1552806936}
            ]
        },
        {
            "id": "000002",
            "key_frame": "3.jpg",
            "status": 2,
            "frames": [
                {"frame_name": "1.jpg", "gps_time": 1555300555},
                {"frame_name": "2.jpg", "gps_time": 1555300560},
                {"frame_name": "3.jpg", "gps_time": 1555300565},
                {"frame_name": "4.jpg", "gps_time": 1555300570},
                {"frame_name": "5.jpg", "gps_time": 1555300580}
            ]
        }
    ]
}

Evaluation Metric

Submissions are scored with a weighted F1‑Score over the three traffic‑state classes. Higher F1 indicates better classification performance.

Expert Judges

The jury includes professors and senior researchers from Peking University, the Chinese Academy of Sciences, and Alibaba’s Amap technical committee, ensuring a rigorous scientific review.

Schedule & Participation

Registration: July 8 – August 28 (UTC+8)

Preliminary round: July 8 – August 31

Semifinal round: September 4 – October 13

Final round: Late October (date TBD)

The dataset becomes publicly downloadable on July 8. After 10:00 AM on July 20, participants may submit JSON results for the test set.

Open to individuals, universities, research institutes, enterprises, and maker teams. Teams may consist of up to three members.

Prizes

Champion: ¥60 000 + certificate

Runner‑up: ¥40 000 + certificate

Third place: ¥20 000 + certificate

Honorable mentions (2 teams): ¥10 000 each + certificate

Top‑10 semifinal teams may receive fast‑track recruitment opportunities with Alibaba Amap.

Communication

Participants can join the DingTalk group (ID 31160357) for announcements and Q&A. QR codes are provided in the original announcement.