Spherical Criteria for Fast and Accurate 360° Object Detection

Authors: Pengyu Zhao, Ansheng You, Yuanxing Zhang, Jiaying Liu, Kaigui Bian, Yunhai Tong12959-12966

AAAI 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental To validate the design of spherical criteria and Reprojection R-CNN, we construct two unbiased synthetic datasets for training and evaluation. Experimental results reveal that compared with the existing criteria, the two-stage detector with spherical criteria achieves the best m AP results under the same inference speed, demonstrating that the spherical criteria can be more suitable for 360 object detection.
Researcher Affiliation Academia School of EECS, Peking University, Beijing, China {pengyuzhao, youansheng, longo, liujiaying, bkg, yhtong}@pku.edu.cn
Pseudocode No The paper describes methods textually and with diagrams, but no structured pseudocode or algorithm blocks are provided.
Open Source Code No The paper does not provide any links to open-source code for the described methodology or state that the code is publicly released.
Open Datasets Yes VOC360 is a synthetic dataset generated from PASCAL VOC 2007 and 2012 (Everingham et al. 2010)... For the multi-object scenario, we construct COCO-Men dataset, which combines the real-world background 360 images and the segmented images of people cropped from COCO dataset (Lin et al. 2014)... To demonstrate the capability of Rep R-CNN for real scenes, we use SUN360 dataset (Xiao et al. 2012).
Dataset Splits Yes VOC360 has 15000 training images, 1800 validation images, and 4955 test images.
Hardware Specification No Both Sph RPN and Rep Net are trained on 4 GPUs for 20 epochs... (No specific GPU model, CPU, or other hardware details are provided).
Software Dependencies No The paper mentions using VGG-16 as a backbone network and refers to concepts like Faster R-CNN, but it does not specify any software names with version numbers (e.g., PyTorch 1.x, TensorFlow 2.x, Python 3.x).
Experiment Setup Yes The learning rate is initially set to 0.001 and then decreased by a factor of 10 after training 15 epochs. The batch size is 16 in Sph RPN and 128 in Rep Net.