Research map / Video / Video segmentation and tracking
Pixel Objectness: Learning to Segment Generic Objects Automatically in Images and Videos
Bo Xiong, Suyog Dutt Jain, Kristen Lorraine Grauman
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 changed | Where | Measured | Effect |
|---|---|---|---|
| add motion stream to appearance stream (joint model) (ablation) | pixel objectness video model · DAVIS | Avg. IoU: 64.69 -> 72.82 | Improved |
| add appearance stream to motion stream (joint model) (ablation) | pixel objectness video model · DAVIS | Avg. IoU: 60.18 -> 72.82 | Improved |
| add motion stream to appearance stream (joint model) (ablation) | pixel objectness video model · YouTube-Objects | Avg. IoU: 68.57 -> 68.95 | Improved |
| add appearance stream to motion stream (joint model) (ablation) | pixel objectness video model · YouTube-Objects | Avg. IoU: 49.43 -> 68.95 | Improved |
| add motion stream to appearance stream (joint model) (ablation) | pixel objectness video model · SegTrack-v2 | Avg. IoU: 56.88 -> 64.44 | Improved |
| add appearance stream to motion stream (joint model) (ablation) | pixel objectness video model · SegTrack-v2 | Avg. IoU: 53.04 -> 64.44 | Improved |
| learn motion model instead of thresholding optical flow | video segmentation · DAVIS | Avg. IoU: 42.95 -> 60.18 | Improved |
| prune pseudo ground truth with bounding boxes (ablation) | motion training data · 100 human-labeled frames | Jaccard: 70.2 -> 77.8 | Improved |
| evaluate on categories unseen in segmented training (Non-PASCAL) | pixel objectness image model · ImageNet-Localization | BBox-CorLoc: 62.12 -> 60.18 | No clear effect |
| use pixel objectness crop for retrieval instead of full image | VGG retrieval · ILSVRC2012 val | mAP: 0.3082 -> 0.3342 | Improved |
| restrict to object-centric classes for retrieval | pixel objectness retrieval · ILSVRC2012 first 400 classes | mAP: 0.3342 -> 0.4166 | Improved |
| use pixel objectness instead of NLC on SegTrack | video segmentation · SegTrack-v2 | Avg. IoU: 80 -> 64.44 | Hurt |
Closest papers by meaning
- Learning Video Object Segmentation With Visual Memory (ICCV 2017 · similarity 0.90)
- FusionSeg: Learning to Combine Motion and Appearance for Fully Automatic Segmentation of Generic Objects in Videos (CVPR 2017 · similarity 0.90)
- Universal Weakly Supervised Segmentation by Pixel-to-Segment Contrastive Learning (ICLR 2021 · similarity 0.90)
- Unsupervised Object Segmentation in Video by Efficient Selection of Highly Probable Positive Features (ICCV 2017 · similarity 0.89)
- Lucid Data Dreaming for Video Object Segmentation (IJCV 2019 · similarity 0.89)
- Learning Hierarchical Image Segmentation For Recognition and By Recognition (ICLR 2024 · similarity 0.89)
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.
