Research map / Point clouds / Point cloud classification
MVImgNet: A Large-Scale Dataset of Multi-View Images
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Pretrain IBRNet on MVImgNet before finetuning | IBRNet · Real-world 360° Objects | PSNR: 22.18 -> 24.67 | Helped |
| Pretrain IBRNet on MVImgNet before finetuning | IBRNet · Realistic Synthetic 360° Objects | PSNR: 25.49 -> 27.26 | Helped |
| Pretrain MVSNet on MVImgNet before DTU finetuning | JDACS / MVSNet · DTU, 5% training samples | 2mm accuracy: 48.61 -> 50.10 | Helped |
| Mix ImageNet training data with MVImgNet views | ResNet-50 · MVImgNet test set | accuracy: 53.09% -> 77.31% | Helped |
| Mix ImageNet training data with MVImgNet views | ResNet-50 · MVImgNet test set | Confidence Var: 0.207 -> 0.102 | Helped |
| Use five non-consecutive frames rather than consecutive frames (ablation) | MVI-Mix image classification · MVImgNet test set | accuracy: 77.31% vs 77.23% | No clear effect |
| Replace multi-view frames with one frame and four data augmentations (ablation) | MVI-Mix image classification · MVImgNet test set | accuracy: 77.31% -> 71.48% | Hurt |
| Finetune MoCo-v2 using same-video frames as positive pairs | MoCo-v2 · MVImgNet test set | accuracy: 70.26% -> 71.22% | Helped |
| Finetune U2Net on MVImgNet with optical-flow consistency loss | U2Net · MVImgNet hard views | IoU: 4.1% | Helped |
| Pretrain PointMAE on MVPNet before ScanObjectNN finetuning | PointMAE · ScanObjectNN PB_T50_RS | overall accuracy: 77.34 -> 84.13 | Helped |
| Pretrain PointNet on MVPNet before finetuning | PointNet · ScanObjectNN PB_T50_RS | overall accuracy: 70.63 -> 67.73 | Hurt |
| Increase MVImgNet pretraining data from 10k to 100k videos (ablation) | JDACS / MVSNet · DTU test set | 2mm accuracy: 52.96 -> 58.63 | Helped |
Closest papers by meaning
- MM-Point: Multi-View Information-Enhanced Multi-Modal Self-Supervised 3D Point Cloud Understanding (AAAI 2024 · similarity 0.88)
- ImageNet3D: Towards General-Purpose Object-Level 3D Understanding (NeurIPS 2024 · similarity 0.88)
- ImOV3D: Learning Open Vocabulary Point Clouds 3D Object Detection from Only 2D Images (NeurIPS 2024 · similarity 0.88)
- From an Image to a Scene: Learning to Imagine the World from a Million 360° Videos (NeurIPS 2024 · similarity 0.87)
- LinNet: Linear Network for Efficient Point Cloud Representation Learning (NeurIPS 2024 · similarity 0.87)
- RRT-MVS: Recurrent Regularization Transformer for Multi-View Stereo (AAAI 2025 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.