Home
Softono
n

nianticlabs

Professional software vendor delivering innovative solutions on the Softono platform. Specialized in both open-source and proprietary software development.

Total Products
2

Software by nianticlabs

monodepth2
Open Source

monodepth2

MonoDepth2 is a reference PyTorch implementation for self-supervised monocular depth estimation, based on the ICCV 2019 paper Digging into Self-Supervised Monocular Depth Prediction by Godard et al. The software enables training and testing models to predict depth maps from single images without requiring ground truth depth data during training. It achieves this through a novel self-supervised approach that leverages stereo pairs or video sequences to minimize photometric reprojection error. The codebase supports various pre-trained models trained with mono-only, stereo-only, or combined mono-stereo modalities at resolutions of 640x192 and 1024x320. Key capabilities include predicting scaled disparity or metric depth for single input images using provided command-line scripts. It supports both training new models and evaluating existing ones on standard benchmarks like KITTI. The system requires PyTorch, CUDA, and OpenCV, with specific recommendations for environment setup using Anaconda to manage dependencie

ML Frameworks
4.5K Github Stars
simplerecon
Open Source

simplerecon

SimpleRecon is a PyTorch implementation for 3D reconstruction and Multi-View Stereo (MVS) depth estimation, introduced in the ECCV 2022 paper SimpleRecon: 3D Reconstruction Without 3D Convolutions. Developed by researchers from institutions including Niantic and the University of Oxford, this software generates dense depth maps for a target image given a set of posed RGB images with known camera intrinsics and extrinsics. Its primary innovation is eliminating computationally expensive 3D convolutions, relying instead on efficient 2D convolutions and cost volume construction to achieve state-of-the-art accuracy. The package supports training and testing on standard datasets such as ScanNetv2 and COLMAP. It includes pre-trained models for immediate inference, tools for evaluating metrics like absolute difference, squared relative error, delta accuracy, Chamfer distance, and F-Score, as well as utilities for point cloud fusion and mesh generation. The codebase is designed for academic research and non-commercial

Design & Creative ML Frameworks
1.4K Github Stars