Cross-Domain 3D Equivariant Image Embeddings
Authors: Carlos Esteves, Avneesh Sud, Zhengyi Luo, Kostas Daniilidis, Ameesh Makadia
ICML 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We show that learning a rich embedding for images with appropriate geometric structure is sufficient for tackling varied applications, such as relative pose estimation and novel view synthesis, without requiring additional task-specific supervision. ... Nonetheless, our promising experimental results indicate our crossdomain embeddings may be useful for a variety of tasks. |
| Researcher Affiliation | Collaboration | *Work done during an internship at Google. 1GRASP Laboratory, University of Pennsylvania 2Google Research. |
| Pseudocode | No | No structured pseudocode or algorithm blocks are present in the paper. Figure 2 shows an architecture diagram, not pseudocode. |
| Open Source Code | No | The paper does not contain an unambiguous statement of code release or a direct link to a source code repository for the described methodology. It mentions 'See supplementary material for more details' in reference to methods comparison, but not code release. |
| Open Datasets | Yes | We utilize the standardized large datasets of 3D shapes Model Net40 (Wu et al., 2015) and Shape Net (Chang et al., 2015) for most of our experiments. ... Our experiments with real images are limited to the airplane and cars categories of Object Net3D (Xiang et al., 2016). |
| Dataset Splits | Yes | For training, we render views in arbitrary poses sampled from SO(3). ... We have two modes of evaluation: instance and category based. For category-based, we measure the relative pose error between each instance and 3 randomly sampled instances from the test set. For instance-based, we measure the error between each instance from the test set and 3 randomly rotated versions of itself. |
| Hardware Specification | No | The paper does not provide specific hardware details (e.g., exact GPU/CPU models, memory amounts, or detailed computer specifications) used for running its experiments. |
| Software Dependencies | No | The paper does not provide specific ancillary software details with version numbers (e.g., library or solver names with version numbers) needed to replicate the experiment. |
| Experiment Setup | Yes | Since our target embeddings are unbounded, we found crucial to use a robust loss such as Huber3, and a Huber breakpoint at 1 works well in practice. ... For all our experiments s(x) is a 10 layer residual spherical CNN trained for Model Net40 3D shape classification on 64 64 inputs. |