Fig. 5: Using MAMMAL for mouse and dog motion capture.
From: Three-dimensional surface motion capture of multiple freely moving pigs using MAMMAL
a An illustration of the video data used for reconstruction. Six views were used. b The mouse mesh model used by MAMMAL. It contains 140 driven joints (left), 14522 vertices (middle), and 22 keypoints (right). c Visualized comparison between DANNCE-T and MAMMAL. d The position of 8 extremities of mouse labeled for quantitative comparison. e, f Box plot of reconstruction errors of different methods on the 8 extremities. Total 50 evenly distributed frames were manually labeled for evaluation (e). Percentage of correctly tracked keypoints versus different thresholds (f). g An illustration of the laboratory environment of Beagle dog social motion capture. Ten GoPro cameras were mounted round the experiment area where two beagle dogs freely moved. h The dog mesh model used by MAMMAL. It contains 39 driven joints (left), 4653 vertices (middle), and 29 keypoints (right). i A 3D rendering of reconstructed dog poses. j, k Box plot of reconstruction errors of MAMMAL and VoxelPose on different view numbers (n = 2 animals, n = 91, 366, 368, 351 and 138 landmarks for Body, FrontLeg, HindLeg, Head and Tail respectively) (j). Percentage of correctly tracked keypoints versus different thresholds (k). In e and j, black bar, median; box shoulders, interquartile range (IQR); whiskers, 1.5 times the IQR; black square dots, mean.
