Bootstrap Your Object Detector via Mixed Training

Authors: Mengde Xu, Zheng Zhang, Fangyun Wei, Yutong Lin, Yue Cao, Stephen Lin, Han Hu, Xiang Bai

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

Reproducibility Variable Result LLM Response
Research Type Experimental We introduce Mix Training, a new training paradigm for object detection that can improve the performance of existing detectors for free. ... In particular, the performance of Faster R-CNN [24] with a Res Net-50 [13] backbone is improved from 41.7 m AP to 44.0 m AP, and the accuracy of Cascade-RCNN [1] with a Swin-Small [22] backbone is raised from 50.9 m AP to 52.8 m AP. ... 4 Experiments
Researcher Affiliation Collaboration 1Huazhong University of Science and Technology 2Xi an Jiaotong University 3Microsoft Research Asia
Pseudocode No The paper includes mathematical equations (1) and (2) but does not contain any structured pseudocode or algorithm blocks.
Open Source Code No The paper does not provide any explicit statements about releasing source code or links to a code repository for the methodology described.
Open Datasets Yes We validate our method on the COCO2017 dataset [20]... [20] Lin, T.-Y., Maire, M., Belongie, S., Hays, J., Perona, P., Ramanan, D., Dollár, P., and Zitnick, C. L. (2014). Microsoft coco: Common objects in context. In ECCV.
Dataset Splits Yes We validate our method on the COCO2017 dataset [20], which contains 80 object categories, 118k images for training (train2017), 5k images (minival) for validation and 20k images for testing (test-dev).
Hardware Specification Yes All the models run on 32 Nvidia V100.
Software Dependencies No The paper mentions optimizers like SGD and AdamW but does not provide specific version numbers for any software libraries, frameworks, or programming languages used (e.g., PyTorch, Python version).
Experiment Setup Yes In our experiments, two different data augmentation strategies are adopted: normal augmentation and strong augmentation. We summarize the details in Table 1. ... For updating the EMA model that predicts pseudo boxes, the momentum coefficient is set to 0.999. ... g is considered an easy target if its score is greater than a threshold (0.9 by default). ... For other training settings and hyper-parameters for each detector, we following the default settings if not otherwise specified.