PFDet: 2nd Place Solution to Open Images Challenge 2018 Object Detection Track

Akiba, Takuya; Kerola, Tommi; Niitani, Yusuke; Ogawa, T.; Sano, Shotaro; Suzuki, Shuji

doi:10.48550/arxiv.1809.00778

Cited by 7 publications

(14 citation statements)

References 10 publications

(18 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…We guess that they are incompatible with the mAP metric in object detection as mentioned in 4.1.2. Our method also outperforms dist-CE loss [24] and Co-BCE loss [1]. The impacts of concurrent softmax during testing.…”

Section: Concurrent Softmaxmentioning

confidence: 79%

“…We propose a soft-balance method together with a hybrid training scheduler to mitigate the over-fitting on infrequent categories and better exploit data of frequent categories. Our methods yield a total gain of 3.34 points, leading to a 60.90 mAP single model result on the public test-challenge set of Open Images, which is 5.09 points higher than the single model result of the first place method [1] on the public test-challenge set last year. More importantly, our overall system achieves a 67.17 mAP, which is 4.29 points higher than their ensembled results.…”

Section: Introductionmentioning

confidence: 83%

“…However, most works are still exploring in datasets like Pascal VOC and MS COCO, which are small by modern standards. Only a few works [25,38,1,10] have been proposed to deal with large-scale object detection dataset like Open Images [16]. Wu et al [38] proposes a soft boxsampling method to cope with the partial label problem in large scale dataset.…”

Section: Related Workmentioning

confidence: 99%

“…A widely used technique, i.e., class-aware sampling mentioned in [31,10,26] is a naive solution to handle the class imbalance problem, in which categories are sampled uniformly in each batch. The class-aware sampling probability P a of class i becomes P a (i) = 1 C . Yet this may cause heavy over-fitting on infrequent categories and insufficient training on frequent categories as aforementioned.…”

Section: Soft-balance Sampling With Hybrid Trainingmentioning

confidence: 99%

See 3 more Smart Citations

Large-Scale Object Detection in the Wild from Imbalanced Multi-Labels

Peng

Sun

et al. 2020

Preprint

View full text Add to dashboard Cite

Training with more data has always been the most stable and effective way of improving performance in deep learning era. As the largest object detection dataset so far, Open Images brings great opportunities and challenges for object detection in general and sophisticated scenarios. However, owing to its semi-automatic collecting and labeling pipeline to deal with the huge data scale, Open Images dataset suffers from label-related problems that objects may explicitly or implicitly have multiple labels and the label distribution is extremely imbalanced. In this work, we quantitatively analyze these label problems and provide a simple but effective solution. We design a concurrent softmax to handle the multi-label problems in object detection and propose a softsampling methods with hybrid training scheduler to deal with the label imbalance. Overall, our method yields a dramatic improvement of 3.34 points, leading to the best single model with 60.90 mAP on the public object detection test set of Open Images. And our ensembling result achieves 67.17 mAP, which is 4.29 points higher than the best result of Open Images public test 2018.

show abstract

Section: Concurrent Softmaxmentioning

confidence: 79%

Section: Introductionmentioning

confidence: 83%

Section: Related Workmentioning

confidence: 99%

Section: Soft-balance Sampling With Hybrid Trainingmentioning

confidence: 99%

See 2 more Smart Citations

Large-Scale Object Detection in the Wild from Imbalanced Multi-Labels

Peng

Sun

et al. 2020

Preprint

View full text Add to dashboard Cite

show abstract

“…Training object detection model from scratch in a reasonable amount of time on a large dataset requires significant resources. For instance, one of the leading models on the Open Images 2018 Object Detection Track needed 33 hours of training on 512 GPUs [1]. Instead of running multiple costly and time consuming training experiments, we focus speeding up optimization with limited hardware resources.…”

Section: Object Detection With Partial Weight Transfermentioning

confidence: 99%

Learning Effective Visual Relationship Detector on 1 GPU

Lu,

Chang,

Rai

et al. 2019

Preprint

View full text Add to dashboard Cite

We present our winning solution to the Open Images 2019 Visual Relationship challenge. This is the largest challenge of its kind to date with nearly 9 million training images. Challenge task consists of detecting objects and identifying relationships between them in complex scenes. Our solution has three stages, first object detection model is finetuned for the challenge classes using a novel weight transfer approach. Then, spatio-semantic and visual relationship models are trained on candidate object pairs. Finally, features and model predictions are combined to generate the final relationship prediction. Throughout the challenge we focused on minimizing the hardware requirements of our architecture. Specifically, our weight transfer approach enables much faster optimization, allowing the entire architecture to be trained on a single GPU in under two days. In addition to efficient optimization, our approach also achieves superior accuracy winning first place out of over 200 teams, and outperforming the second place team by over 5% on the held-out private leaderboard.

show abstract

Key Indicators to Assess the Performance of LiDAR-Based Perception Algorithms: A Literature Review

Karri,

da Silva,

Correia

2023

IEEE Access

View full text Add to dashboard Cite

Perception algorithms are essential for autonomous or semi-autonomous vehicles to perceive the semantics of their surroundings, including object detection, panoptic segmentation, and tracking. Decision-making in case of safety-critical situations, like autonomous emergency braking and collision avoidance, relies on the outputs of these algorithms. This makes it essential to correctly assess such perception systems before their deployment and to monitor their performance when in use. It is difficult to test and validate these systems, particularly at runtime, due to the high-level and complex representations of their outputs. This paper presents an overview of different existing metrics used for the evaluation of LiDAR-based perception systems, emphasizing particularly object detection and tracking algorithms due to their importance in the final perception outcome. Along with generally used metrics, we also discuss the impact of Planning KL-Divergence (PKL), Timed Quality Temporal Logic (TQTL), and Spatio-temporal Quality Logic (STQL) metrics on object detection algorithms. In the case of panoptic segmentation, Panoptic Quality (PQ) and Parsing Covering (PC) metrics are analysed resorting to some pretrained models. Finally, it addresses the application of diverse metrics to evaluate different pretrained models with the respective perception algorithms on publicly available datasets. Besides the identification of the various metrics being proposed, their performance and influence on models are also assessed after conducting new tests or reproducing the experimental results of the reference under consideration.

show abstract

PFDet: 2nd Place Solution to Open Images Challenge 2018 Object Detection Track

Cited by 7 publications

References 10 publications

Large-Scale Object Detection in the Wild from Imbalanced Multi-Labels

Large-Scale Object Detection in the Wild from Imbalanced Multi-Labels

Learning Effective Visual Relationship Detector on 1 GPU

Key Indicators to Assess the Performance of LiDAR-Based Perception Algorithms: A Literature Review

Contact Info

Product

Resources

About