Large-Scale Unsupervised Object Discovery
Authors: Van Huy Vo, Elena Sizikova, Cordelia Schmid, Patrick Pérez, Jean Ponce
NeurIPS 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments on COCO [42] and Open Images [35] show that, in the single-object discovery setting where a single prominent object is sought in each image, the proposed LOD (Large-scale Object Discovery) approach is on par with, or better than the state of the art for mediumscale datasets (up to 120K images), and over 37% better than the only other algorithms capable of scaling up to 1.7M images. |
| Researcher Affiliation | Collaboration | 1INRIA, Département d informatique de l ENS, ENS, CNRS, PSL University, Paris, France 2Valeo.ai 3Center for Data Science, New York University |
| Pseudocode | No | The paper does not contain structured pseudocode or algorithm blocks. |
| Open Source Code | Yes | Our code is publicly available at https://github.com/huyvvo/LOD. |
| Open Datasets | Yes | We consider two large public datasets: C120K, a combination of all images in the training and validation sets of the COCO 2014 dataset [42], except those contain only crowd objects, with approximately 120,000 images depicting 80 object classes and Open Images (Op1.7M) [35], the largest dataset ever evaluated for UOD so far, with 1.7 million images. |
| Dataset Splits | No | The paper mentions using a 'combination of all images in the training and validation sets' of COCO for C120K and also uses subsets like C20K and Op50K for evaluation. However, it does not explicitly provide the specific percentages or sample counts for training, validation, or test splits for its own experiments on these datasets, nor does it refer to predefined splits in a way that allows reproduction of the splitting. |
| Hardware Specification | No | In our experiments, we use 4,000 CPUs for preprocessing for all methods, and 48 CPUs for the optimization step in [32] and LOD, the maximum possible with the Mat Lab parallel toolbox used in our implementation. The paper specifies the number of CPUs but does not provide specific CPU models or other detailed hardware specifications (e.g., GPU models, memory). |
| Software Dependencies | No | The paper mentions 'Mat Lab parallel toolbox used in our implementation' but does not specify the version numbers for MATLAB or the toolbox, or any other software dependencies with their versions. |
| Experiment Setup | Yes | For optimization, we choose β = 10 4 in (P) and α = 10% in LOD. To select objects from ranked proposals in an image, we choose proposal i as an object if it has the highest score in the image or the intersection over union (Io U) between i and each of the previously selected object regions is at most 0.3. |