HEAP: Unsupervised Object Discovery and Localization with Contrastive Grouping
Authors: Xin Zhang, Jinheng Xie, Yuan Yuan, Michael Bi Mi, Robby T. Tan
AAAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experimental results on semantic segmentation retrieval, unsupervised object discovery, and saliency detection tasks demonstrate that HEAP achieves state-of-the-art performance. |
| Researcher Affiliation | Collaboration | 1National University of Singapore 2Huawei International Pte Ltd |
| Pseudocode | No | The paper describes the method with mathematical formulas and textual descriptions, but it does not include a clearly labeled pseudocode or algorithm block. |
| Open Source Code | No | The paper does not provide an explicit statement or link for the open-source code of the described methodology. |
| Open Datasets | Yes | We extensively evaluate the proposed HEAP on semantic segmentation retrieval on VOC12 (Everingham et al. 2012), unsupervised saliency detection on DUT-OMRON (Yang et al. 2013), DUTS-TE (Wang et al. 2017), and ECSSD (Shi et al. 2015), and unsupervised object discovery on VOC07 & VOC12 (Everingham et al. 2007, 2012), and COCO20k (Lin et al. 2014). |
| Dataset Splits | No | The paper mentions using a 'val set' for semantic segmentation retrieval: 'We build the feature bank on the train split and find the nearest neighbors of each object in the val set by retrieving the feature bank and assign them the corresponding ground-truth labels.' However, it does not provide specific details on how this split was generated or its size for reproducibility in the main text. |
| Hardware Specification | No | The paper does not provide specific hardware details (e.g., exact GPU/CPU models or processor types) used for running its experiments. |
| Software Dependencies | No | The paper mentions using a 'frozen pre-trained encoder' like DINO and Dense CRF for post-processing, but it does not specify any version numbers for these or other software dependencies. |
| Experiment Setup | No | The paper states 'Implementation details are provided in the supplementary material' and discusses general aspects of the model and training parameters like the number of group tokens (M) and the weight (α) for the inter-image clustering loss, but it does not provide specific hyperparameter values or detailed training configurations within the main text. |