Instance Mining with Class Feature Banks for Weakly Supervised Object Detection
Authors: Yufei Yin, Jiajun Deng, Wengang Zhou, Houqiang Li3190-3198
AAAI 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments conducted on two publicly available datasets, Pascal VOC 2007 and 2012, demonstrate the effectiveness of our method. |
| Researcher Affiliation | Academia | 1 CAS Key Laboratory of GIPAS, EEIS Department, University of Science and Technology of China 2 Institute of Artificial Intelligence, Hefei Comprehensive National Science Center |
| Pseudocode | Yes | Algorithm 1: CFB updating strategy |
| Open Source Code | No | The paper does not provide concrete access to source code (specific repository link, explicit code release statement, or code in supplementary materials) for the methodology described. |
| Open Datasets | Yes | We evaluate our proposed method on both Pascal VOC 2007 and Pascal VOC 2012 (Everingham et al. 2010) following previous WSOD works. |
| Dataset Splits | Yes | We train on trainval split (5,011 images for VOC 2007, 11,540 images for VOC 2012) |
| Hardware Specification | Yes | Our experiments are implemented based on Py Torch on NVIDIA GTX 1080Ti GPUs. |
| Software Dependencies | No | The paper mentions "Py Torch" but does not specify a version number or any other software dependencies with version numbers. |
| Experiment Setup | Yes | The batch size, momentum, and weight decay are set to 4, 0.9, and 5 10 4 respectively. The learning rate is set to 1 10 3 for the first 60K iterations and 1 10 4 for the following 20K iterations. For data augmentation, we use five image scales, i.e., {480, 576, 688, 864, 1200} for the shortest side of images, and random horizontal flipping is applied. We set K = 6 for the VOC 2007 and K = 8 for the VOC 2012. We set τ = 0.5 for both datasets. Considering that the stability of the network increases during the training process, we set α = 5 at the first 40K iterations and then tighten the restriction with α = 2. The number of refinement branches T is set to 2. |