Our goal is to learn a deep network that, given a small number of images of an object of a given category, reconstructs it in 3D. While several recent works have obtained analogous results using synthetic data or assuming the availability of 2D primitives such as keypoints, we are interested in working with challenging real data and with no manual annotations. We thus focus on learning a model from multiple views of a large collection of object instances. We contribute with a new large dataset of object centric videos suitable for training and benchmarking this class of models. We show that existing techniques leveraging meshes, voxels, or implicit surfaces, which work well for reconstructing isolated objects, fail on this challenging data. Finally, we propose a new neural network design, called warp-conditioned ray embedding (WCR), which significantly improves reconstruction while obtaining a detailed implicit representation of the object surface and texture, also compensating for the noise in the initial SfM reconstruction that bootstrapped the learning process. Our evaluation demonstrates performance improvements over several deep monocular reconstruction baselines on existing benchmarks and on our novel dataset.
Our method takes as input an image and produces per pixel features using a U-Net $\Phi$. We then shoot rays from a target view and retrieve per-pixel features from one or multiple source images. Once all spatial feature vectors are aggregated into a single feature vector, we combine them with their harmonic embeddings and pass them to an MLP yielding per location colors and opacities. Finally, we use differentiable raymarching to produce a rendered image.
For the purpose of studying learning 3D object categories in the wild, we crowd-sourced a large collection of videos from AMT.