JUMPS: Joints Upsampling Method for Pose Sequences

Machine Learning Approaches for 3D Motion Synthesis and Musculoskeletal Dynamics Estimation: A Survey

“…The recognized classes can be then used to train the generator in a supervised manner. Similarly, Mourot et al [54] present a method that estimates complex movements, such as jumps of a 2D avatar, by incorporating in the architecture a GAN and an encoder that help to learn mappings from human pose sequences to GAN's latent space. This method has the ability to upsample the number of joints in each pose sequence in order to correct any missing or occluded joints.…”

“…[11] NSM with local motion phases Own Mocap Dataset Real-Time (Starke, 2021) [12] NSM with control modules Own Mocap Dataset Real-Time and Offline (Henter, 2020) [37] LSTM-based CMU [107] and HDM05 [108] and dog Mocap from [43] Real-time (Peng, 2021) [41] Adversarial RL Commercial Mocap dataset Real-time (Starke, 2022) [42] Periodic Autoencoder Various Mocap Datasets (e.g. AIST++, dog mocap [43]) Real-time (Aristidou, 2021) [45] DL methods like LSTM Own Mocap Dataset and AIST++ Real-time (Pu, 2022) [46] Transformer-based AIST++ [59] Offline (Men, 2021) [53] GAN (LSTM-based) SBU [109] and HHOI [110] and 2C [111] Real-Time (Mourot, 2020) [54] GAN-based MPI-INF-3DHP [112] Offline (Wang J., 2021) [48] GAN-based PROX [113], GTA-IM [114] Offline (Li P., 2022) [49] GAN-based -Offline (Hassan, 2021) [50] Autoregressive CVAE-based Own Mocap dataset Offline (Ling, 2020) [52] VAE + RL Own Mocap Dataset Real-time (Li J., 2021) [55] HM-VAE AMASS, 3DPW, LAFAN1 Offline (Cai, 2021) [56] CVAE H3.6M, CMU Offline (Briq, 2022) [57] RTVAE-Multi PROX, Charades [115] Offline (Raab, 2023) [61] Transformer-based Diffusion Model HumanML3D [116] and Mixamo [117] and more Offline (Ma J., 2022) [62] Transformer-based Diffusion Model AMASS [104], LAFAN1…”