Depth Privileged Object Detection in Indoor Scenes via Deformation Hallucination

Authors: Zhijie Zhang, Yan Liu, Junjie Chen, Li Niu, Liqing Zhang3456-3464

AAAI 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experimental results on NYUDv2 and SUN RGB-D demonstrate the effectiveness of our method against the state-of-the-art methods for depth privileged object detection.
Researcher Affiliation Academia Mo E Key Lab of Artificial Intelligence, Department of Computer Science and Engineering, Shanghai Jiao Tong University {zzj506506, loseover, chen.bys, ustcnewly}@sjtu.edu.cn, zhang-lq@cs.sjtu.edu.cn
Pseudocode No The paper describes its methods in text and diagrams, but does not include any structured pseudocode or algorithm blocks.
Open Source Code No The paper does not provide any explicit statement about releasing its source code or a link to a code repository.
Open Datasets Yes NYU Depth V2 (NYUDv2) (Silberman et al. 2012) consists of 1449 paired RGB-D images. SUN RGB-D (Song, Lichtenberg, and Xiao 2015) is composed of an official train/test split with 5285 and 5050 images, respectively.
Dataset Splits Yes The dataset [NYUDv2] is split into training (795 images) and test (654 images) sets. SUN RGB-D (Song, Lichtenberg, and Xiao 2015) is composed of an official train/test split with 5285 and 5050 images, respectively.
Hardware Specification Yes All experiments are conducted on Ubuntu 18.04 with two 8GB Ge Force RTX 2080 SUPER, 16GB Intel 9700K, and Py Torch 1.2.0 on Python 3.7.
Software Dependencies Yes All experiments are conducted on Ubuntu 18.04 with two 8GB Ge Force RTX 2080 SUPER, 16GB Intel 9700K, and Py Torch 1.2.0 on Python 3.7.
Experiment Setup Yes We train our model using the SGD optimizer for 50k iterations for D-branch pre-training and the whole model training. The basic learning rate is initialized to 1 10 3 and reduced to 1 10 4 when the iterations reach 40k. The weight decay and momentum are set to 5 10 4 and 0.9, respectively. The random seed is set to 222. two trade-off parameters α and β are set as 1.0 and 2.0, respectively. δ is a hyper-parameter controlling the intensity of avoiding negative transfer and set as 0.25 via cross-validation. µ is a trade-off parameter and set as 0.1 via corss-validation.