Research map / Detection and segmentation / 3D detection and driving
Deep Exemplar 2D-3D Detection by Adapting From Real to Rendered Views
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 |
|---|---|---|---|
| Adapt real-image features to align with rendered-view features (ablation) | 2D-3D chair detection on Pascal VOC · Pascal VOC chair subset | AP: 17.9 -> 52.3 | Helped |
| Add a ReLU after the fully-connected adaptation layer (ablation) | 2D-3D exemplar detection · IKEA instance detection, original annotations | mAP: 11.94 -> 12.38 | Helped |
| Add score calibration to exemplar-view similarities (ablation) | 2D-3D chair detection · Pascal VOC chair subset | AP: 5.6 -> 52.3 | Helped |
| Use fully-connected adaptation rather than limited-support convolution (ablation) | Feature adaptation for chair detection · Pascal VOC chair subset | AP | Helped |
| Use one adaptation layer instead of two (ablation) | Feature adaptation for chair detection · Pascal VOC chair subset | AP | Helped |
| Add ReLU after the convolutional adaptation layer (ablation) | Feature adaptation for chair detection · Pascal VOC chair subset | AP | Helped |
| Use cosine distance rather than L2 distance or dot product for pool4 retrieval (ablation) | CaffeNet image-to-CAD retrieval · IKEA instance retrieval | accuracy: 57.7 / 47.5 -> 65.0 | Helped |
| Increase the number of CAD rendered views (ablation) | Fully-connected projection · Pascal VOC chair subset | AP: 33.3 -> 50.0 | Helped |
| Increase diversity of CAD models in the library (ablation) | Fully-connected projection · Pascal VOC chair subset | AP: 21.7 -> 50.0 | Helped |
| Use real-image compositing in synthetic positives instead of white backgrounds (ablation) | Logistic pool4 detector without adaptation · Pascal VOC chair subset | AP: 1.4 -> 3.7 | Helped |
| Use feature adaptation rather than composited synthetic positives alone (ablation) | Exemplar detector · Pascal VOC chair subset | AP: 36.4 -> 52.3 | Helped |
| Use fc7 rather than pool4 features for logistic regression (ablation) | Adapted logistic classifier · Pascal VOC chair subset | AP: 12.9 -> 26.6 | Helped |
Closest papers by meaning
- Omni3D: A Large Benchmark and Model for 3D Object Detection in the Wild (CVPR 2023 · similarity 0.85)
- HVPR: Hybrid Voxel-Point Representation for Single-Stage 3D Object Detection (CVPR 2021 · similarity 0.85)
- 3D Bounding Box Estimation Using Deep Learning and Geometry (CVPR 2017 · similarity 0.84)
- Multi-Level Fusion Based 3D Object Detection From Monocular Images (CVPR 2018 · similarity 0.84)
- MV2Cyl: Reconstructing 3D Extrusion Cylinders from Multi-View Images (NeurIPS 2024 · similarity 0.84)
- ImOV3D: Learning Open Vocabulary Point Clouds 3D Object Detection from Only 2D Images (NeurIPS 2024 · similarity 0.83)
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.