Feature-Critic Networks for Heterogeneous Domain Generalization
Authors: Yiying Li, Yongxin Yang, Wei Zhou, Timothy Hospedales
ICML 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experimental evaluation demonstrates that our method outperforms state-of-the-art solutions in both settings. Our evaluation shows good performance in the conventional DG setting using Rotated MNIST (Ghifary et al., 2015; Motiian et al., 2017) and PACS (Li et al., 2017a) benchmarks, as well as the heterogeneous DG setting using the larger scale Visual Decathlon (VD) (Rebuffi et al., 2017) benchmark. 4. Experiments |
| Researcher Affiliation | Collaboration | 1College of Computer, National University of Defense Technology, Hunan, China 2School of Informatics, The University of Edinburgh, Edinburgh, UK 3Samsung AI Centre, Cambridge, UK. |
| Pseudocode | Yes | Algorithm 1 Simulating Domain Shift in Training; Algorithm 2 Full Algorithm |
| Open Source Code | Yes | Our demo code can be viewed on https://github.com/liyiying/Feature_Critic. |
| Open Datasets | Yes | The Visual Decathlon dataset, initially proposed for multi-domain learning (Rebuffi et al., 2017), also provides a large scale and rigorous benchmark for DG. Rotated MNIST (Ghifary et al., 2015) contains six domains with each corresponding to a degree of roll rotation in the classic MNIST dataset. PACS (Li et al., 2017a) is a recent object recognition benchmark for domain generalisation. |
| Dataset Splits | Yes | To realise our idea, we simulate training-to-testing domain shift by splitting our source domains into virtual training and testing (i.e., validation) domains. Randomly split D: Dtrn Dval = Dtrn Dval = D. During each iteration, we randomly choose four of the six source domains as meta-train, and the remaining two provide the meta-test (validation) domains. We follow the standard protocol and perform leave-one-domain-out evaluation. |
| Hardware Specification | No | The paper does not provide specific hardware details (e.g., GPU models, CPU types, memory) used for the experiments. It mentions 'NVIDIA Corporation GPU donation' but no specific models or configurations. |
| Software Dependencies | No | The paper mentions optimizers like AMSGrad and M-SGD but does not specify software versions for libraries, frameworks (e.g., PyTorch, TensorFlow), or programming languages used. |
| Experiment Setup | Yes | We train all components end-to-end using the AMSGrad (Reddi et al., 2018) (batch-size/per meta-train domain=64, batch-size/per meta-test domain=32, lr=0.0005, weight decay=0.0001) for 30k iterations where the lr decayed in 5K, 12K, 15K, 20K iterations by a factor 5, 10, 50, 100, respectively. For Feature-Critic, we train using the AMSGrad optimizer (lr=0.001, weight decay=0.00005) for 5,000 iterations. Our Feature-Critic (set embedding variant) is trained with M-SGD optimizer (batch size/per meta-trian domain=32, batch size/per meta-test domain=16, lr=0.0005, weight decay=0.00005, momentum=0.9) for 45K iterations. |