Towards Unsupervised Model Selection for Domain Adaptive Object Detection
Authors: Hengfu Yu, Jinhong Deng, Wen Li, Lixin Duan
NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We have conducted extensive experiments on various DAOD benchmarks and approaches, and the experimental results show that the proposed DAS correlates well with the performance of DAOD models and can be used as an effective tool for model selection after training. |
| Researcher Affiliation | Academia | Hengfu Yu Jinhong Deng Wen Li Lixin Duan University of Electronic Science and Technology of China hfyu@std.uestc.edu.cn, {jhdengvision, liwenbnu, lxduan}@gmail.com |
| Pseudocode | No | The paper describes the methods mathematically and textually but does not include any explicitly labeled pseudocode or algorithm blocks. |
| Open Source Code | Yes | The code will be released at https://github.com/Henry Yu23/DAS. |
| Open Datasets | Yes | Benchmarks. We follow previous works [11, 50, 38] and evaluate the effectiveness of the proposed DAS on the following adaptation scenarios: Real-to-Art Adaptation (P2C): In this scenario, we test our proposed method with domain shift between the real image domain and the artistic image domain. Following [50], we choose the PASCAL VOC 2007/2012 and Clipart1k as the source and target domains, respectively. ... Weather Adaptation (C2F): ... we use the Cityscapes and Foggy Cityscapes as the source and target domains, respectively. ... Synthetic-to-Real Adaptation (S2C): ... we utilize Sim10k as the source domain and Cityscapes as the target domain. |
| Dataset Splits | Yes | Cityscapes [13] contains a diverse set of urban street scenes captured from 50 cities and 2, 975 training images and 500 validation images, annotated for 8 object classes. |
| Hardware Specification | Yes | Training is conducted on four RTX 3090 GPUs or two A100 GPUs according to the computational requirements of different DAOD methods. |
| Software Dependencies | No | The paper mentions "Our implementation is built upon the Detectron2 detection framework" but does not specify a version number for Detectron2 or any other key software libraries like Python, PyTorch, or CUDA. |
| Experiment Setup | Yes | The hyperparameters, learning rates, and optimizers are set according to the default configurations provided in the original papers. We set our hyperparameters λ = 1 and γ = 1 in all experiments, which perform well on all the benchmarks. |