Mind the Gap: Polishing Pseudo Labels for Accurate Semi-supervised Object Detection
Authors: Lei Zhang, Yuxuan Sun, Wei Wei
AAAI 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experiments on both PASCAL VOC and MS-COCO benchmarks demonstrate the superiority of the proposed method over existing state-of-the-art baselines. |
| Researcher Affiliation | Academia | 1 School of Computer Science, Northwestern Polytechnical University, China 2 Research & Development Institute of Northwestern Polytechnical University in Shenzhen, China {nwpuzhanglei,weiweinwpu}@nwpu.edu.cn, sunyuxuan@mail.nwpu.edu.cn |
| Pseudocode | No | The paper includes architectural diagrams (Figure 2, Figure 3) but does not provide any structured pseudocode or algorithm blocks. |
| Open Source Code | Yes | The code can be found at https://github.com/snowdusky/DualPolishLearning. |
| Open Datasets | Yes | PASCAL VOC: We employ images from the training-validation set trainval in PASCAL VOC07 as the labeled data, while images from the training-validation set trainval in PASCAL VOC12 as the unlabeled data... MS-COCO: Similar as (Sohn et al. 2020b; Zhou et al. 2021), we separately randomly select 1%, 5% and 10% images from the COCO training set train2017 as the labeled data while the remaining are used as unlabeled data to evaluate the SSOD performance under different amounts of labeled data. |
| Dataset Splits | Yes | PASCAL VOC: We employ images from the training-validation set trainval in PASCAL VOC07 as the labeled data, while images from the training-validation set trainval in PASCAL VOC12 as the unlabeled data... MS-COCO: Similar as (Sohn et al. 2020b; Zhou et al. 2021), we separately randomly select 1%, 5% and 10% images from the COCO training set train2017 as the labeled data while the remaining are used as unlabeled data to evaluate the SSOD performance under different amounts of labeled data. |
| Hardware Specification | No | The paper does not provide specific hardware details (e.g., exact GPU/CPU models, memory amounts) used for running its experiments. |
| Software Dependencies | No | The hyper-parameters in detectors are determined according to the MMDetection toolbox (Chen et al. 2019). The paper mentions the MMDetection toolbox but does not provide specific version numbers for it or any other software dependencies. |
| Experiment Setup | Yes | Due to limited space, the SSOD training details as well as the hyper-parameters (e.g., θclsc, Nclsc, θclsm, Nclsm, τpos etc.) setting for the proposed method are given in the supplementary material. ... Table 6: Effect of different hyperparameters (a) Effect of θreg (b) Effect of γ (c) Effect of η |