Unsupervised Part Discovery from Contrastive Reconstruction
Authors: Subhabrata Choudhury, Iro Laina, Christian Rupprecht, Andrea Vedaldi
NeurIPS 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Our method yields semantic parts which are consistent across fine-grained but visually distinct categories, outperforming the state of the art on three benchmark datasets. Code is available at the project page: https://www.robots.ox.ac.uk/~vgg/research/unsup-parts/. In the following we validate our approach on three benchmark datasets, the Caltech-UCSD Birds-200 dataset (CUB-200-2011) [74], the large-scale fashion database (Deep Fashion) [54] and PASCALPart [12]. Details regarding the datasets are given in the appendix. We carry out ablation experiments to study (a) the importance of the proposed objective functions, and (b) the role of supervised vs. unsupervised pre-training for the different components of our model. Lastly, we show that our method compares favorably to prior work both quantitatively and qualitatively. |
| Researcher Affiliation | Academia | Subhabrata Choudhury Iro Laina Christian Rupprecht Andrea Vedaldi Visual Geometry Group University of Oxford Oxford, UK subha,iro,chrisr,vedaldi@robots.ox.ac.uk |
| Pseudocode | No | The paper describes its proposed approach and objective functions but does not include any formal pseudocode or algorithm blocks. |
| Open Source Code | Yes | Code is available at the project page: https://www.robots.ox.ac.uk/~vgg/research/unsup-parts/. |
| Open Datasets | Yes | In the following we validate our approach on three benchmark datasets, the Caltech-UCSD Birds-200 dataset (CUB-200-2011) [74], the large-scale fashion database (Deep Fashion) [54] and PASCALPart [12]. |
| Dataset Splits | No | The paper mentions using datasets for training and validation but does not explicitly state the specific train/validation/test splits, percentages, or methodology for creating them in the main text. |
| Hardware Specification | No | The paper does not specify any particular hardware used for running the experiments (e.g., GPU models, CPU types, or cloud computing instances). |
| Software Dependencies | No | The paper mentions network architectures like 'Deep Lab-v2', 'Res Net-50', and 'VGG19' but does not provide specific version numbers for any software dependencies (e.g., Python, PyTorch, TensorFlow, CUDA versions). |
| Experiment Setup | No | The paper states, 'We use the same set of hyper-parameters for both, CUB-200 and Deep-Fashion, whereas some small changes are necessary for PASCAL-Part since the images are in a different resolution which typically impacts the magnitude of feature-based losses. We provide all implementation details in the appendix.' However, specific hyperparameter values (e.g., learning rate, batch size, number of epochs) are not explicitly detailed in the main text. |