Almost all software and data is for research purposes only. Please, check the corresponding licenses.
NOTE: Some code and data can also be found under the link Code/Data in publications.


Combining Implicit Function Learning and Parametric Models for 3D Human Reconstruction
Combining Implicit Function Learning and Parametric Models for 3D Human Reconstruction. ECCV'20(Oral)
  • Pre-trained IPNet models to predict the full dressed surface of human, body shape under clothing and SMPL part correspondences.
  • Code for training and testing IPNet.
  • Code to register SMPL/SMPL+D to IPNet predictions.
  • Codes for data pre-processing.
Code/Data


Neural Articulated Shape Approximation
Neural Articulated Shape Approximation. ECCV'20
  • Dataset.
  • Code for training and testing NASA.
Code/Data


SIZER: A Dataset and Model for Parsing 3D Clothing and Learning Size Sensitive 3D Clothing
SIZER: A Dataset and Model for Parsing 3D Clothing and Learning Size Sensitive 3D Clothing. ECCV'20(Oral)
  • 3D scans and SMPL+G registrations of 100 subjects in various garments styles and sizes in A-pose.
  • Code for training and testing ParserNet and SizerNet.
  • Pre-trained ParserNet and SizerNet models.
Code/Data


Unsupervised Shape and Pose Disentanglement for 3D Meshes
Unsupervised Shape and Pose Disentanglement for 3D Meshes. ECCV'20
  • Pre-trained disentangled auto-encoder on AMASS that can infer latent shape and pose codes from human bodies.
  • Code for preprocessing and training on AMASS.
Code/Data


TailorNet: Predicting Clothing in 3D as a Function of Human Pose, Shape and Garment Style
TailorNet: Predicting Clothing in 3D as a Function of Human Pose, Shape and Garment Style. CVPR'20(Oral)
  • Dataset of 3D garments in various pose, shape and garment styles.
  • Code for training and inference of TailorNet model.
  • Pretrained TailorNet model for multiple garments.
Code/Data


Implicit Functions in Feature Space for 3D Shape Reconstruction and Completion
Implicit Functions in Feature Space for 3D Shape Reconstruction and Completion. CVPR'20
  • Pre-trained models for: Point Cloud Completion, Voxel-Superresolution and Human Single-View Reconstruction
  • Code for: data processing, IF-Net model, training, generation and evaluation.
  • Link to the ShapeNet data used.
Code/Data


Learning to Transfer Texture from Clothing Images to 3D Humans
Learning to Transfer Texture from Clothing Images to 3D Humans. CVPR'20
  • Code and pre-trained models for mapping the front and back of clothing images - TShirts, Pants, ShortPants - downloaded from clothing stores to 3D garment meshses worn atop the SMPL body model.
  • Code for optimization method to obtain training data.
  • Code for training the neural model - Pix2Surf.
Code/Data


Multi-Garment Net: Learning to Dress 3D People from Images
Multi-Garment Net: Learning to Dress 3D People from Images. ICCV'19
  • Released 'Digital Wardrobe' of registered garments to dress SMPL model. The dataset includes, segmented 3D scans, registrations, garments and texture maps.
  • Code to dress SMPL model with the released garments.
  • Code to retarget clothing across subjects with different body shapes and poses.
  • Pre-trained MGN model to infer 3D garments and underlying body shape from a few RGB images.
Code/Data


Tex2Shape: Detailed Full Human Body Geometry from a Single Image
Tex2Shape: 3D avatars from a single image. ICCV'19
  • Code and model for the pose-independent image-to-image translation method to reconstruct a full-body avatar from a single image. The avatars feature fine details even on unseen parts!
  • Reconstructions are based on the SMPL body topology so they can be re-posed and re-shaped.
Code/Data


Learning to Reconstruct People in Clothing from a Single RGB Camera
Octopus: 3D avatars from a video. CVPR'19
  • Code for the learning+optimization based method to reconstruct a detailed body shape including clothing and hair from a single RGB video. Given semantic segmentations as input, reconstructions are computed in 10 seconds!
  • Reconstructions are based on the SMPL body topology so they can be re-posed and re-shaped.
Code/Data


Detailed Human Avatars from Monocular Video
Sharp texture map from multiple views/frames. 3DV'18
  • Code. INPUT: Multiple images (frames, or views) and a 3D human model roughly aligned to them. OUTPUT: texture map, sharp and without stitching artefacts.
Code/Data


Video Based Reconstruction of 3D People Models
3D avatars from a video and PeopleSnapshot Dataset. CVPR'18
  • Code for the optimization based method to reconstruct a detailed body shape including clothing and hair from a single RGB video. Reconstructions are based on the SMPL body topology so they can be re-posed and re-shaped.
  • PeopleSnapshot dataset including 24 subjects rotating in front of the camera.
Code/Data


Deep Inertial Poser: Learning to Reconstruct Human Pose from Sparse Inertial Measurements in Real Time
Deep Inertial Poser: Learning to Reconstruct Human Pose from Sparse Inertial Measurements in Real Time. SiggraphAsia'18.
  • Code for real time 3D pose from 6 IMUs. Training and testing parts along the real-time demo.
  • DIP-IMU: large scale dataset including real and syntehtic IMU readings paired with SMPL poses. Largest IMU dataset available.
Code/Data


Neural Body Fitting: Unifying Deep Learning and Model Based Human Pose and Shape Estimation
Neural Body Fitting. Human Body Shape and Pose from a Single Image. 3DV'18.
  • Code for 3D human pose and shape estimation from a single image. Better performance than SOTA when trained with the same data.
Code/Data


Recovering Accurate 3D Human Pose in The Wild Using IMUs and a Moving Camera
3DPW: 3D Poses in the Wild Dataset. ECCV'18.
Unique dataset with monocular hand-held video together with accurate 3D human poses for evaluation. Very accurate 3D poses are obtained combining video and IMU.
  • 60 video sequences: downtown, taking the bus, having coffee, sports, etc.
  • 2D pose annotations
  • 3D poses reference poses
  • Camera poses for every frame in the sequences
  • 3D body scans and 18 3D human models with different clothing variations.
Code/Data


Single-Shot Multi-Person 3D Pose Estimation From Monocular RGB
Multiperson 3D pose dataset
  • MuPoTS-3D: Dataset of multiple people with reference 3D poses. Compared to 3DPW above, the camera is static but reference pose for more than 2 persons is provided
  • MuCo-3DHP: Scripts to generate a syntehtic 3D multiple people dataset by compositing pictures.
Code/Data


A Generative Model of People in Clothing
Image Based Generative Model of People in Clothing. ICCV'17
Based on conditional variational autoencoder.
  • Models: trained models. They allow to generate semantic segmentation images of people and to "translate" a SMPL rendering into a photo-realistic image.
  • Training code (based on TensorFlow v1.1).
  • The training dataset is built on top of Chictopia10K. We provide processed annotations as well as the SMPL body model fit to the images.
Code/Data


Detailed, accurate, human shape estimation from clothed 3D scan sequences
BUFF dataset.
BUFF a unique dataset of 4D dynamic scans of people in clothing.
  • High quality ~9.000 4D scans of people in clothing
  • Ground truth 3D shape under clothing.
  • 5 subjects, 3 male and 2 female
  • 2 clothing styles: a) t-shirt and long pants and b) a soccer outfit.
  • 3 different motions i) hips ii) tilt_twist_left iii) shoulders_mill.
Code/Data


Dynamic FAUST: Registering Human Bodies in Motion
DFAUST.
  • This dataset is a unique resource containing over 40,000 4D scans of multiple people; 4D means 3D scans over time. Processing 4D data is challenging, so we provide aligned data in which we have registered a common template mesh to all scans. This alignment process takes into account geometry and surface texture to make it accurate. The dataset includes the raw scan data, registered template meshes, and masks that say where the template mesh is sufficiently accurate to be considered ground truth.
Code/Data


Human Pose Estimation from Video and IMUs
Multimodal Motion Capture Dataset (TNT15). PAMI'16
  • video data: multi-view sequences obtained from 8 calibrated RGB-cameras.
  • silhouettes: binary segmented images obtained by background subtraction.
  • IMU data: orientation and acceleration data of 10 IMUs.
  • projection matrices: camera parameters of all 8 cameras.
  • meshes: 3D laser scans and registered meshes of each actor.
Code/Data


Dyna: A Model of Dynamic Human Shape in Motion
Dyna dataset. SIGGRAPH'15
  • The dataset contains the 40.000 registered dynamic meshes used to train Dyna. Sequences exhibit dynamic soft-tissue deformations.
Code/Data


SMPL: A Skinned Multi-Person Linear Model
SMPL model. SIGGRAPH Asia'15
  • 3D body model of pose and shape. Learned from thousands of scans.
  • Useful for computer vision. Body shape, 2D/3D pose, etc.
  • Compatible with most graphics packages (FBX). Scripts to run it in Python, Maya, Unity.
  • Regular updates: dynamic blendshapes, model improvements, etc.
Code/Data