Lucid Data Dreaming for Video Object Segmentation
arXiv:1703.09554
Abstract
Convolutional networks reach top quality in pixel-level video object segmentation but require a large amount of training data (1k~100k) to deliver such results. We propose a new training strategy which achieves state-of-the-art results across three evaluation datasets while using 20x~1000x less annotated data than competing methods. Our approach is suitable for both single and multiple object segmentation. Instead of using large training sets hoping to generalize across domains, we generate in-domain training data using the provided annotation on the first frame of each video to synthesize ("lucid dream") plausible future video frames. In-domain per-video training data allows us to train high quality appearance- and motion-based models, as well as tune the post-processing stage. This approach allows to reach competitive results even when training from only a single annotated frame, without ImageNet pre-training. Our results indicate that using a larger training set is not automatically better, and that for the video object segmentation task a smaller training set that is closer to the target domain is more effective. This changes the mindset regarding how many training samples and general "objectness" knowledge are required for the video object segmentation task.
arXiv commentAccepted in International Journal of Computer Vision (IJCV)
References in corpus (18)
- Rethinking Atrous Convolution for Semantic Image Segmentation
- Efficient Inference in Fully Connected CRFs with Gaussian Edge Potentials
- Learning from Synthetic Humans
- Modeling and Propagating CNNs in a Tree Structure for Visual Tracking
- FusionSeg: Learning to combine motion and appearance for fully automatic segmention of generic objects in videos
- Siamese Instance Search for Tracking
- PixelNet: Representation of the pixels, by the pixels, and for the pixels
- Learning Video Object Segmentation from Static Images
- Video Object Segmentation with Re-identification
- RefineNet: Multi-Path Refinement Networks for High-Resolution Semantic Segmentation
- Guided Optical Flow Learning
- Video Propagation Networks
- Learning to Segment Instances in Videos with Spatial Propagation Network
- Learning Features by Watching Objects Move
- Click Carving: Segmenting Objects in Video with Point Clicks
- Super-Trajectory for Video Segmentation
- Video Object Segmentation using Tracked Object Proposals
- FOMTrace: Interactive Video Segmentation By Image Graphs and Fuzzy Object Models
Cited by in corpus (17)
- Video Object Segmentation with Re-identification
- RVOS: End-to-End Recurrent Network for Video Object Segmentation
- Interactive Video Object Segmentation in the Wild
- MaskRNN: Instance Level Video Object Segmentation
- MHP-VOS: Multiple Hypotheses Propagation for Video Object Segmentation
- DMM-Net: Differentiable Mask-Matching Network for Video Object Segmentation
- Modular Interactive Video Object Segmentation: Interaction-to-Mask, Propagation and Difference-Aware Fusion
- A Transductive Approach for Video Object Segmentation
- Learning to Generate Synthetic Data via Compositing
- Proposal, Tracking and Segmentation (PTS): A Cascaded Network for Video Object Segmentation
- Discriminative Online Learning for Fast Video Object Segmentation
- OVSNet : Towards One-Pass Real-Time Video Object Segmentation
- Realistic Hair Simulation Using Image Blending
- VideoMatch: Matching based Video Object Segmentation
- Fast and Accurate Online Video Object Segmentation via Tracking Parts
- Temporally Coherent Video Harmonization Using Adversarial Networks
- Design Pseudo Ground Truth with Motion Cue for Unsupervised Video Object Segmentation