VisDrone-DET2021: The Vision Meets Drone Object detection Challenge Results

You Only Look Once (YOLO) series detectors are suitable for aerial image object detection because of their excellent real-time ability and performance. Their high performance depends heavily on the anchor generated by clustering the training set. However, the effectiveness of the general Anchor Generation algorithm is limited by the unique data distribution of the aerial image dataset. The divergence in the distribution of the number of objects with different sizes can cause the anchors to overfit some objects or be assigned to suboptimal layers because anchors of each layer are generated uniformly and affected by the overall data distribution. In this paper, we are inspired by experiments under different anchors settings and proposed the Layered Anchor Generation (LAG) algorithm. In the LAG, objects are layered by their diagonals, and then anchors of each layer are generated by analyzing the diagonals and aspect ratio of objects of the corresponding layer. In this way, anchors of each layer can better match the detection range of each layer. Experiment results showed that our algorithm is of good generality that significantly uprises the performance of You Only Look Once version 3 (YOLOv3), You Only Look Once version 5 (YOLOv5), You Only Learn One Representation (YOLOR), and Cascade Regions with CNN features (Cascade R-CNN) on the Vision Meets Drone (VisDrone) dataset and the object DetectIon in Optical Remote sensing images (DIOR) dataset, and these improvements are cost-free.

show abstract

“…The effectiveness of our algorithm is tested on two public aerial image benchmarks, DIOR [ 41 ] and VisDrone [ 42 ].…”

Section: Experiments and Resultsmentioning

confidence: 99%

LAG: Layered Objects to Generate Better Anchors for Object Detection in Aerial Images

Wan

Tan

et al. 2022

Sensors

View full text Add to dashboard Cite

show abstract

“…We used the VisDrone-DET2021 dataset [ 51 ], which was used in the VisDrone-DET2021 Challenge, an ICCV 2021 workshop. The same dataset has been used previously in the VisDrone2019 [ 52 ] challenge.…”

Section: Methodsmentioning

confidence: 99%

YOLOv5 with ConvMixer Prediction Heads for Precise Object Detection in Drone Imagery

Baidya

Jeong

2022

Sensors

View full text Add to dashboard Cite

The potency of object detection techniques using Unmanned Aerial Vehicles (UAVs) is unprecedented due to their mobility. This potency has stimulated the use of UAVs with object detection functionality in numerous crucial real-life applications. Additionally, more efficient and accurate object detection techniques are being researched and developed for usage in UAV applications. However, object detection in UAVs presents challenges that are not common to general object detection. First, as UAVs fly at varying altitudes, the objects imaged via UAVs vary vastly in size, making the task at hand more challenging. Second due to the motion of the UAVs, there could be a presence of blur in the captured images. To deal with these challenges, we present a You Only Look Once v5 (YOLOv5)-like architecture with ConvMixers in its prediction heads and an additional prediction head to deal with minutely-small objects. The proposed architecture has been trained and tested on the VisDrone 2021 dataset, and the acquired results are comparable with the existing state-of-the-art methods.

show abstract

“…However, there is no ball-related dataset in the literature. We adopts harmonization-aware image compositing [37] to generate synthetic but realistic balls on pedestrian datasets, e.g., VisDrone [3], COCO ball-person subset, PANDA [58]. To address the efficiency, we design a lightweight model based on YOLOv3 and YOLOv5.…”

Section: System Designmentioning

confidence: 99%

“…Last, the dilated gated signal is eroded by θ/2. we mix PANDA [58], VisDrone pedestrian subset [3], and COCO ball-person subset [36] with LPCVC21-train to obtain a large-scale mixed dataset for ball-person detection, dubbed as MixDet-train. MixDet-test is the testing set of LPCV21, namely LPCVC21-test.…”

Section: Robustnessmentioning

confidence: 99%

E^2TAD: An Energy-Efficient Tracking-based Action Detector

Hu¹,

Wu²,

Miao³

et al. 2022

Preprint

View full text Add to dashboard Cite

Video action detection (spatio-temporal action localization) is usually the starting point for human-centric intelligent analysis of videos nowadays. It has high practical impacts for many applications across robotics, security, healthcare, etc. The two-stage paradigm of Faster R-CNN inspires a standard paradigm of video action detection in object detection, i.e., firstly generating person proposals and then classifying their actions. However, none of the existing solutions could provide fine-grained action detection to the "who-when-where-what" level. This paper presents a tracking-based solution to accurately and efficiently localize predefined key actions spatially (by predicting the associated target IDs and locations) and temporally (by predicting the time in exact frame indices). This solution won first place in the UAV-Video Track of 2021 Low-Power Computer Vision Challenge (LPCVC).

show abstract

VisDrone-DET2021: The Vision Meets Drone Object detection Challenge Results

Cited by 109 publications

References 23 publications

LAG: Layered Objects to Generate Better Anchors for Object Detection in Aerial Images

LAG: Layered Objects to Generate Better Anchors for Object Detection in Aerial Images

YOLOv5 with ConvMixer Prediction Heads for Precise Object Detection in Drone Imagery

E^2TAD: An Energy-Efficient Tracking-based Action Detector

Contact Info

Product

Resources

About