FourierFormer: Enhancing Clutter-Free Room Modeling with Fourier Transform
Authors : Bruno Henriques, Benjamin Allaert, Jean-Philippe Vandeborre
Keywords : Inpainting, Equirectangular Vision, Room Modeling
We propose an innovative approach for inpainting indoor environments from a single image, addressing the limitations of current methods that rely on structure estimators. With our FourierMixer block and an aggressive masking policy, our unified network effectively captures the periodic structures of indoor spaces, enhancing scene reconstruction with symmetry, particularly horizon/ceiling lines and cuboid-shaped rooms.
eMotion-GAN: A Motion-based GAN for Photorealistic and Facial Expression Preserving Frontal View Synthesis
Authors : Omar Ikne, Benjamin Allaert, Ioan Marius Bilasco, Hazem Wannous
Keywords : Facial expressions, frontal view synthesis, optical flow, motion warping, generative adversarial networks
We propose an innovative approach to facial expression recognition (FER) by mitigating the impact of head pose variations. Unlike existing frontalization methods that often introduce distortions, eMotion-GAN accurately preserves facial expressions by filtering out motion caused by head variations, which is considered noise, while retaining motion related to expressions.
AG-MAE: Anatomically Guided Spatio-Temporal Masked Auto-Encoder for Online Hand Gesture Recognition
Authors : Omar Ikne, Benjamin Allaert, Hazem Wannous
Keywords : Hand gesture, Anatomical constraints, Mask Auto-Encoder
AG-MAE is a novel approach that integrates anatomical constraints to guide the self-supervised training of a spatio-temporal masked autoencoder. This enhances the learning of 3D keypoint representations, making gesture recognition more accurate and suitable for online and continuous applications.