Hot start! The upload channel for the preliminary round of the first round of the Zhouyi Contest has been opened, come and see your ranking!

Questioner : OPEN AI LAB

Scenario : Multi-person scene face capture machine product

Competition rules : Participants provide face detection algorithms to detect multiple faces in a single image (more than 30 faces). Faces may be partially occluded. The algorithm must be run on EAIDK. The winner is the one with the highest accuracy and fastest speed.

Competition process :

Registration and preliminary round time: May 10th - May 20th (teams ranked 10% will enter the finals) - Starting from May 10th, you can log in to the official website of the competition to upload your entries. The deadline for uploading works is May 20th. The preliminary round will be evaluated online. The model will be converted into Tengine format and uploaded to the server. The test set will be run in the background and the evaluation results will be fed back.

Finals time: May 26 (top three) - The finals will be conducted offline. The model will be deployed on EAIDK610 and tested using Tengine, and the evaluation results will be fed back.

Finals venue: 3rd Floor, Building B, Student Innovation Center, Shanghai Jiao Tong University (800 Dongchuan Road, Minhang District, Shanghai)

prize settings :

First prize: 50,000 yuan for team 1

Second Prize: 15,000 yuan for team

Third Prize: 5,000 RMB for 1 team

Evaluation Rules :

The competition does not specify a training set, and the contestants are responsible for solving the problem themselves. The test set consists of 6,000 images, each with a resolution of 1280×720, including at least 30 faces from different angles. The faces in the test set include different genders, ages, and skin colors. The face annotation images are as follows:

The model uses two indicators, face detection recall and model calculation speed, to evaluate, with weights of 0.75 and 0.25 respectively. The specific calculation formula is:

Score = 0.15*(R1+R2+R3+R4+R5) + 0.25*Ts/Tx

Ts represents the shortest time taken by the algorithm among all participating teams

Tx represents the time taken by each team's algorithm.

R1-R5 represent the recall values ​​under the number of false positives of 100, 200, 500, 1000, and 2000 respectively.

The calculation method of face recall rate is as follows:

Among them, True positive represents the number of faces detected, True positive + False negative

Indicates the number of faces in the entire sample

In order to ensure the fairness of the algorithm, algorithm teams with Recall1 lower than 0.5 will be eliminated.

Applicants:

Enterprises, technical makers, and students can all participate in the competition. Each team consists of 1 to 5 people.