DETReg: Unsupervised Pretraining with Region Priors for Object Detection

Bar, Amir; Wang, Xin; Kantorov, Vadim; Reed, Colorado; Herzig, Roei; Chechik, Gal; Rohrbach, Anna; Darrell, Trevor; Globerson, Amir

doi:10.1109/cvpr52688.2022.01420

Cited by 83 publications

(35 citation statements)

References 40 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Supervised fine-tuning. In addition to evaluating the self-method AP 50 AP 75 AP AR 1 AR 10 AR 100 UP-DETR [18] 0.0 0.0 0.0 0.0 0.0 0.4 Selective Search [67] 0.5 0.1 0.2 0.2 1.5 10.9 DETReg [68] 3 supervised instance segmenter directly, we also evaluate the performance of our approach in a supervised setting by fine-tuning the self-supervised instance segmenter with annotations. As shown in…”

Section: Resultsmentioning

confidence: 99%

FreeSOLO: Learning to Segment Objects without Annotations

Wang

Yu²,

Mello³

et al. 2022

2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

View full text Add to dashboard Cite

show abstract

Section: Resultsmentioning

confidence: 99%

FreeSOLO: Learning to Segment Objects without Annotations

Wang

Yu²,

Mello³

et al. 2022

2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

View full text Add to dashboard Cite

show abstract

“…LodeSTAR builds on geometric deep learning 13 and the recent surge of self-supervised object tracking methods [14][15][16][17][18][19][20][21] to create a self-supervised (or more precisely, self-distillative) object-detection neural network optimized for microscopy data. Specifically, we exploit the fact that a neural network that is equivariant to rotations and translations (i.e., a neural network for which a roto-translational transformation of the input image produces an equivalent roto-translation of the prediction) operates as an object detector (see Methods, "Theory of geometric self-distillation").…”

Section: Lodestar Overviewmentioning

confidence: 99%

“…It marks cells as found if a predicted position overlaps with the segmentation of the cell. Using this method, we evaluate the F1-score of LodeSTAR, as well as those of four other self-supervised object detection methods: SoCo 18 , FSDet 21 , InstanceLoc 20 , and DETReg 19 . See Methods, "Object detection comparison", for more details.…”

Section: Validation With Experimental Datamentioning

confidence: 99%

See 1 more Smart Citation

Single-shot self-supervised object detection in microscopy

et al. 2022

View full text Add to dashboard Cite

Object detection is a fundamental task in digital microscopy, where machine learning has made great strides in overcoming the limitations of classical approaches. The training of state-of-the-art machine-learning methods almost universally relies on vast amounts of labeled experimental data or the ability to numerically simulate realistic datasets. However, experimental data are often challenging to label and cannot be easily reproduced numerically. Here, we propose a deep-learning method, named LodeSTAR (Localization and detection from Symmetries, Translations And Rotations), that learns to detect microscopic objects with sub-pixel accuracy from a single unlabeled experimental image by exploiting the inherent roto-translational symmetries of this task. We demonstrate that LodeSTAR outperforms traditional methods in terms of accuracy, also when analyzing challenging experimental data containing densely packed cells or noisy backgrounds. Furthermore, by exploiting additional symmetries we show that LodeSTAR can measure other properties, e.g., vertical position and polarizability in holographic microscopy.

show abstract

“…However, unlike CutLER, such salient object detection methods only locate a single, usually the most prominent, object and cannot be used for real world images containing multiple objects. While some recent methods, e.g., FreeSOLO [47] and DETReg [3], also aim at unsupervised multi-object detection (or multi-object discovery), they rely on a particular detection architecture, e.g., SOLO-v2 [48] or DDETR [5,54]. Additionally, apart from self-supervised features trained on ImageNet [12], the current state-of-theart methods FreeSOLO and MaskDistill [42] also require 'in-domain' unlabeled data for model training.…”

Section: Introductionmentioning

confidence: 99%

Cut and Learn for Unsupervised Object Detection and Instance Segmentation

Wang¹,

Girdhar²,

Yu³

et al. 2023

Preprint

View full text Add to dashboard Cite

We propose Cut-and-LEaRn (CutLER), a simple approach for training unsupervised object detection and segmentation models. We leverage the property of selfsupervised models to 'discover' objects without supervision and amplify it to train a state-of-the-art localization model without any human labels. CutLER first uses our proposed MaskCut approach to generate coarse masks for multiple objects in an image, and then learns a detector on these masks using our robust loss function. We further improve performance by self-training the model on its predictions. Compared to prior work, CutLER is simpler, compatible with different detection architectures, and detects multiple objects. CutLER is also a zero-shot unsupervised detector and improves detection performance AP 50 by over 2.7× on 11 benchmarks across domains like video frames, paintings, sketches, etc. With finetuning, CutLER serves as a lowshot detector surpassing MoCo-v2 by 7.3% AP box and 6.6% AP mask on COCO when training with 5% labels.

show abstract

DETReg: Unsupervised Pretraining with Region Priors for Object Detection

Cited by 83 publications

References 40 publications

FreeSOLO: Learning to Segment Objects without Annotations

FreeSOLO: Learning to Segment Objects without Annotations

Single-shot self-supervised object detection in microscopy

Cut and Learn for Unsupervised Object Detection and Instance Segmentation

Contact Info

Product

Resources

About