🤖 AI Summary
High-fidelity, joint-articulable 3D foot modeling under no/weak supervision remains challenging. Method: We propose Foot Implicit Neural Deformation (FIND), an end-to-end framework integrating implicit neural representations (INRs) and neural deformation fields, enabling disentangled reconstruction of shape, texture, and joint pose from a single monocular 2D image. We introduce a novel unsupervised part-level loss and establish a progressive learning paradigm—from zero-label initialization to weakly supervised refinement—to enforce anatomical consistency and articulation awareness. Additionally, we release Foot3D, the first high-accuracy 3D foot dataset. Results: On Foot3D, FIND significantly outperforms PCA-based baselines in shape fidelity and part correspondence accuracy. The unsupervised loss improves robustness of 2D inference under occlusion and viewpoint variation. Furthermore, FIND supports arbitrary-resolution mesh extraction, demonstrating strong potential for multi-platform deployment.
📝 Abstract
In this paper we present a high fidelity and articulated 3D human foot model. The model is parameterised by a disentangled latent code in terms of shape, texture and articulated pose. While high fidelity models are typically created with strong supervision such as 3D keypoint correspondences or pre-registration, we focus on the difficult case of little to no annotation. To this end, we make the following contributions: (i) we develop a Foot Implicit Neural Deformation field model, named FIND, capable of tailoring explicit meshes at any resolution i.e. for low or high powered devices; (ii) an approach for training our model in various modes of weak supervision with progressively better disentanglement as more labels, such as pose categories, are provided; (iii) a novel unsupervised part-based loss for fitting our model to 2D images which is better than traditional photometric or silhouette losses; (iv) finally, we release a new dataset of high resolution 3D human foot scans, Foot3D. On this dataset, we show our model outperforms a strong PCA implementation trained on the same data in terms of shape quality and part correspondences, and that our novel unsupervised part-based loss improves inference on images.
