Key Performance Requirements for Spring 24:
Subsystem | System Criteria | Our Performance |
Drone Control | MSE between drone’s positions and the waypoints <= 15 cm2 | 11.02 cm2 |
Drone Data Streaming | Video streamed from the drone must average 30 frames per second(FPS) | 30 FPS |
Perception | Depth Estimation error on ground truth <= 10cm | 0.5545 cm |
Perception | The 3D pose estimation model should be able to detect all the joints of humans in any position provided without occlusion. RMSE per joint on any dataset should be < 5cm | 4.87 cm |
Avatar | RMSE should be less than 10 cm between the input 3D poses used to control the avatar and the detected 3D poses of the reconstructed avatar, where the camera is fixed. | 4.578 cm |
Subsystem Performance Details
Drone Control
Position of the drone
MSE error on each axis and total
Depth Estimation
Relative Depth Map
Comparison with Grountruth
Avatar Generation Quality
Joint | RMSE |
Joint 0 | 3.130 |
Joint 1 | 5.183 |
Joint 2 | 4.251 |
Joint 3 | 7.385 |
Joint 4 | 4.197 |
Joint 5 | 2.300 |
Joint 6 | 1.903 |
Joint 7 | 2.895 |
Joint 8 | 3.586 |
Joint 9 | 6.117 |
Joint 10 | 4.387 |
Joint 11 | 4.444 |
Joint 12 | 10.059 |
Joint 13 | 7.472 |
Joint 14 | 3.399 |
Joint 15 | 4.312 |
Joint 16 | 2.817 |
Average Results | 4.578 |