OW-DETR: Open-world Detection Transformer

Gupta, Akshita; Narayan, Sanath; Joseph, K J; Khan, Salman; Khan, Fahad Shahbaz; Shah, Mubarak

doi:10.1109/cvpr52688.2022.00902

Cited by 105 publications

(55 citation statements)

References 22 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…We show results for FPN and cascade models trained and evaluated on PACO-LVIS in Tab. 16. Cascade models improve the performance for all but the largest model.…”

Section: E Additional Zero-shot Instance Detection Resultsmentioning

confidence: 95%

“…Merge: After the splitting phase the annotated groups are very coherent, i.e., the majority of occurrences in the same group belong to the same instance. However due to video fragmentation and additional limitation on the number of boxes that can be shown to annotators (16) many occurrence groups belong to the same instance and need to be merged. To address this we use similarity in DINO model [2] embedding space.…”

Section: A3 Annotation Pipeline A31 Instance Annotationmentioning

confidence: 99%

“…Today, tasks requiring fine-grained understanding of objects like open vocabulary detection [10,16,22,55], GQA [19], and referring expressions [4,23,34] are gaining importance besides traditional object detection. Representing objects through category labels is no longer sufficient.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

PACO: Parts and Attributes of Common Objects

Ramanathan¹,

Anmol²,

Petrović³

et al. 2023

Preprint

View full text Add to dashboard Cite

show abstract

“…We show results for FPN and cascade models trained and evaluated on PACO-LVIS in Tab. 16. Cascade models improve the performance for all but the largest model.…”

Section: E Additional Zero-shot Instance Detection Resultsmentioning

confidence: 95%

Section: A3 Annotation Pipeline A31 Instance Annotationmentioning

confidence: 99%

See 1 more Smart Citation

PACO: Parts and Attributes of Common Objects

Ramanathan¹,

Anmol²,

Petrović³

et al. 2023

Preprint

View full text Add to dashboard Cite

show abstract

“…Pseudo labels play an important role in guiding models to detect unknown object instances, determining the upper learning limitation of the model. The existing methods [13,17] only use model-driven pseudo-labelling and do not take full advantage of the inputs' priori knowledge (light flow, textures, 𝑒𝑡𝑐). The model-driven pseudo-labelling [13] makes the model's learning get caught up in the knowledge of known objects, for the reason that the only source of knowledge for the model is known object instances.…”

Section: Self-adaptive Pseudo-labellingmentioning

confidence: 99%

“…However, its success largely relies on a held-out validation set which is leveraged to estimate the distribution of unknown objects in the energy-based classifier. To alleviate the problems in ORE, OW-DETR [13] proposes to use the detection transformer [3,38] for OWOD in a justifiable way and directly leverages the framework of DDETR [38]. In addition, OW-DETR proposes an attention-driven PLM which selects pseudo labels for unknown objects according to the attention scores.…”

Section: Introductionmentioning

confidence: 99%

CAT: LoCalization and IdentificAtion Cascade Detection Transformer for Open-World Object Detection

Ma¹,

Wang²,

Fan³

et al. 2023

Preprint

View full text Add to dashboard Cite

Open-world object detection (OWOD), as a more general and challenging goal, requires the model trained from data on known objects to detect both known and unknown objects and incrementally learn to identify these unknown objects. The existing works which employ standard detection framework and fixed pseudo-labelling mechanism (PLM) have the following problems: (𝑖) The inclusion of detecting unknown objects substantially reduces the model's ability to detect known ones. (𝑖𝑖) The PLM does not adequately utilize the priori knowledge of inputs. (𝑖𝑖𝑖) The fixed selection manner of PLM cannot guarantee that the model is trained in the right direction. We observe that humans subconsciously prefer to focus on all foreground objects and then identify each one in detail, rather than localize and identify a single object simultaneously, for alleviating the confusion. This motivates us to propose a novel solution called CAT: LoCalization and IdentificAtion Cascade Detection Transformer which decouples the detection process via the shared decoder in the cascade decoding way. In the meanwhile, we propose the self-adaptive pseudo-labelling mechanism which combines the model-driven with inputdriven PLM and self-adaptively generates robust pseudolabels for unknown objects, significantly improving the ability of CAT to retrieve unknown objects. Comprehensive experiments on two benchmark datasets, 𝑖.𝑒., MS-COCO and PASCAL VOC, show that our model outperforms the stateof-the-art in terms of all metrics in the task of OWOD, incremental object detection (IOD) and open-set detection.

show abstract

Generative Unknown-Aware Object Detection

Wang

Chen

et al. 2022

Intelligent Robotics and Applications

View full text Add to dashboard Cite

OW-DETR: Open-world Detection Transformer

Cited by 105 publications

References 22 publications

PACO: Parts and Attributes of Common Objects

PACO: Parts and Attributes of Common Objects

CAT: LoCalization and IdentificAtion Cascade Detection Transformer for Open-World Object Detection

Generative Unknown-Aware Object Detection

Contact Info

Product

Resources

About