Unsupervised Part Representation by Flow Capsules
Authors: Sara Sabour, Andrea Tagliasacchi, Soroosh Yazdani, Geoffrey Hinton, David J Fleet
ICML 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We evaluate Flow Capsules on unsupervised part segmentation and unsupervised image classification. Experiments demonstrate robust part discovery in the presence of multiple objects, cluttered backgrounds, and occlusion. Section 5. Experiments: We evaluate Flow Capsules on images with different dynamics, shapes, backgrounds and textures. Geo. For this synthetic dataset, ... Exercise. This dataset contains natural images... |
| Researcher Affiliation | Collaboration | 1Google Research, Brain Team. 2Department of Computer Science, University of Toronto. |
| Pseudocode | No | The paper does not contain any pseudocode or clearly labeled algorithm blocks. |
| Open Source Code | No | The paper does not provide any explicit statements about open-sourcing code or links to a code repository for the described methodology. |
| Open Datasets | Yes | Geo. For this synthetic dataset, we use the same code and setup as (Xu et al., 2019)... Geo+. This variant of Geo incorporates natural image backgrounds (random images from Image Net (Deng et al., 2009)), and textured foreground shapes. Textures are random samples from the Brodatz dataset (Picard et al., 1993). |
| Dataset Splits | Yes | Geo. For this synthetic dataset, we use the same code and setup as (Xu et al., 2019), generating 100k images for training, 1k for validation, and 10k for testing. |
| Hardware Specification | No | The paper does not provide any specific hardware details (e.g., CPU, GPU models, memory, or cloud instances) used for running the experiments. |
| Software Dependencies | No | The paper mentions the use of "Adam optimizer (Kingma & Ba, 2014)", but it does not specify version numbers for this or any other software libraries, frameworks, or programming languages used. |
| Experiment Setup | Yes | Models are trained using the Adam optimizer (Kingma & Ba, 2014) with a fixed learning rate of 1e 4 for 150 epochs. We use C=32 and K=8 for Geo models and C=16 and K=16 for Exercise model. Regularization constants for Lcenter and Lsmooth are 1e 2 and 1e 4. |