ultralytics/yolov3: Rectangular Inference, Conv2d + Batchnorm2d Layer Fusion

Jocher, Glenn; guigarfr,; Perry,; Ttayu,; Veitch-Michaelis, Josh; Bianconi, Gabriel; Baltacı, Fatih; Suess, Daniel; WannaSeaU,; IlyaOvodov,

doi:10.5281/zenodo.2672652

Cited by 4 publications

(3 citation statements)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…We propose to learn efficient deep object detectors through pruning less important feature channels and further present SlimYOLOv3 with fewer trainable parameters and lower computation overhead for real-time object detection on UAV s. We empirically demonstrate the effectiveness of SlimYOLOv3 on VisDrone2018-Det benchmark dataset [18]. SlimYOLOv3 is implemented based on the publicly available Darknet [16] and a Pytorch implementation for YOLOv3 [30]. We use a Linux server with Intel(R) Xeon(R) E5-2683 v3 CPU @ 2.00GHz (56 CPUs), 64GB RAM, and four NVIDIA GTX1080ti GPU cards to train and evaluate models in our experiments.…”

Section: Methodsmentioning

confidence: 99%

See 1 more Smart Citation

SlimYOLOv3: Narrower, Faster and Better for Real-Time UAV Applications

Zhang¹,

Zhong²,

Li³

2019

2019 IEEE/CVF International Conference on Computer Vision Workshop (ICCVW)

210

120

View full text Add to dashboard Cite

Drones or general Unmanned Aerial Vehicles (UAVs), endowed with computer vision function by onboard cameras and embedded systems, have become popular in a wide range of applications. However, real-time scene parsing through object detection running on a UAV platform is very challenging, due to limited memory and computing power of embedded devices. To deal with these challenges, in this paper we propose to learn efficient deep object detectors through channel pruning of convolutional layers. To this end, we enforce channel-level sparsity of convolutional layers by imposing L1 regularization on channel scaling factors and prune less informative feature channels to obtain "slim" object detectors. Based on such approach, we present SlimYOLOv3 with fewer trainable parameters and floating point operations (FLOPs) in comparison of original YOLOv3 (Joseph Redmon et al., 2018) as a promising solution for real-time object detection on UAVs. We evaluate SlimYOLOv3 on VisDrone2018-Det benchmark dataset; compelling results are achieved by SlimYOLOv3 in comparison of unpruned counterpart, including ~90.8% decrease of FLOPs, ~92.0% decline of parameter size, running ~2 times faster and comparable detection accuracy as YOLOv3. Experimental results with different pruning ratios consistently verify that proposed SlimYOLOv3 with narrower structure are more efficient, faster and better than YOLOv3, and thus are more suitable for real-time object detection on UAVs. Our codes are made publicly available at https://github.com/PengyiZhang/SlimYOLOv3. Figure 1. Billion floating point operations (BFLOPs) versus accuracy (mAP) on VisDrone2018-Det benchmark dataset. Enabled by channel pruning, our SlimYOLOv3-SPP3 achieves comparable detection accuracy as YOLOv3 but only requires the equivalent floating point operations as YOLOv3-tiny. Such performance is very competitive in drone applications. Details are given in §5.

show abstract

Section: Methodsmentioning

confidence: 99%

“…It is to be noted that we use Darknet [16] to perform normal training and fine-tuning, while we use the Pytorch implementation [30] to perform sparsity training for convenience.…”

Section: Trainingmentioning

confidence: 99%

SlimYOLOv3: Narrower, Faster and Better for Real-Time UAV Applications

Zhang¹,

Zhong²,

Li³

2019

2019 IEEE/CVF International Conference on Computer Vision Workshop (ICCVW)

210

120

View full text Add to dashboard Cite

show abstract

“…3) Training details: All experiments were conducted with mmdetection implementations [14] except for YOLOv3 Py-Torch implementation released by Ultralytics LLC [15]. Microsoft COCO dataset [13] pre-trained model is used for the parameter initialization.…”

Section: Methodsmentioning

confidence: 99%

Field-of-View IoU for Object Detection in 360° Images

Chen¹,

Ikehata²,

Aizawa³

2022

Preprint

View full text Add to dashboard Cite

360°cameras have gained popularity over the last few years. In this paper, we propose two fundamental techniques-Field-of-View IoU (FoV-IoU) and 360Augmentation for object detection in 360°images. Although most object detection neural networks designed for the perspective images are applicable to 360°images in equirectangular projection (ERP) format, their performance deteriorates owing to the distortion in ERP images. Our method can be readily integrated with existing perspective object detectors and significantly improves the performance. The FoV-IoU computes the intersection-overunion of two Field-of-View bounding boxes in a spherical image which could be used for training, inference, and evaluation while 360Augmentation is a data augmentation technique specific to 360°object detection task which randomly rotates a spherical image and solves the bias due to the sphere-to-plane projection. We conduct extensive experiments on the 360°indoor dataset with different types of perspective object detectors and show the consistent effectiveness of our method.

show abstract