AI Paper Atlas

Research map / Point clouds / Point cloud classification

Image2Point: 3D Point-Cloud Understanding with 2D Image Pretrained Models

Chenfeng Xu, Shijia Yang, Tomer Galanti, Bichen Wu, Xiangyu Yue, Bohan Zhai, Wei Zhan, Peter Vajda, Kurt Keutzer, Masayoshi Tomizuka

ECCV 2022 · Point clouds › Point cloud classification · approach: Local feature aggregation networks · 72 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (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 changedWhereMeasuredEffect
add ImageNet1K pretraining and fully finetune (FIP-ALL)ResNet152 · ModelNet 3D Warehousetop-1 accuracy (%): 90.28 -> 91.09Improved
use a deeper model with image pretrainingFIP-ALL ImageNet1K · ModelNet 3D Warehousetop-1 accuracy (%): ResNet18 +0.13 vs ResNet152 +0.81Improved
pretrain on ImageNet21K instead of ImageNet1KFIP-ALL ResNet50 · ModelNet 3D Warehousetop-1 accuracy (%): 90.92 -> 91.05Improved
add image pretraining to ViTViT-B-16 / ViT-L-16 · ModelNet 3D Warehousetop-1 accuracy (%): 84.27 -> 87.77; 83.48 -> 87.66Improved
add ImageNet1K pretrainingResNet18 on SemanticKITTI · SemanticKITTImIoU (%): 64.75 -> 65.57Improved
use Cityscapes pretrained HRNetV2-W48 instead of ImageNet1KHRNetV2-W48 · SemanticKITTImIoU (%): +5.25% mIoU over training from scratch (ImageNet1K: 3.41)Improved
pretrain with ImageNet images instead of PointContrastModelNet classification · ModelNet 3D Warehousetop-1 accuracy (%): 89.95 -> 90.88Improved
add ImageNet1K pretraining in 1-shot settingResNet18/50/152 · ModelNet 3D Warehouse 1-shottop-1 accuracy (%): 26.8 -> 36.8 (+10.0)Improved
add ImageNet1K pretraining as teacher in semi-supervised distillationResNet34 · ModelNet 3D Warehouse 1-shottop-1 accuracy (%): 29.2 -> 41.1 (+11.9)Improved
add ImageNet1K pretraining to speed up trainingResNet50 · ModelNet 3D Warehouseepochs to 90% top-1: 122 -> 11 epochs (11.1x)Improved
also finetune BN layers in addition to input and output layers (FIP-IO+BN vs FIP-IO) (ablation)FIP ResNet50 ImageNet1K · ModelNet 3D Warehousetop-1 accuracy (%): 81.20 -> 89.87Improved
also update BN mean and variance (IOms vs IO) (ablation)FIP ResNet50 ImageNet1K · ModelNet 3D Warehousetop-1 accuracy (%): 81.20 -> 82.94Improved

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy