Discovery of Latent 3D Keypoints via End-to-end Geometric Reasoning
Authors: Supasorn Suwajanakorn, Noah Snavely, Jonathan J. Tompson, Mohammad Norouzi
NeurIPS 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We conduct three sets of experiments on different object categories from the Shape Net [6] dataset. We evaluate our technique against a strongly supervised baseline using manually annotated keypoints on the task of relative 3D pose estimation. Surprisingly, we find that our end-to-end framework achieves significantly better results, despite the lack of keypoint annotations. |
| Researcher Affiliation | Collaboration | Supasorn Suwajanakorn Noah Snavely Jonathan Tompson Mohammad Norouzi supasorn@vistec.ac.th, {snavely, tompson, mnorouzi}@google.com Vidyasirimedhi Institute of Science and Technology Google AI |
| Pseudocode | No | The paper describes the computational steps and architecture details but does not include a formally labeled 'Pseudocode' or 'Algorithm' block, nor does it present any structured, code-like procedural steps. |
| Open Source Code | No | The paper states: 'The discovered 3D keypoints on the car, chair, and plane categories of Shape Net [6] are visualized at keypointnet.github.io.' This link is for visualizations and does not provide access to the source code for the methodology described in the paper. There is no explicit statement of code release. |
| Open Datasets | Yes | Our training data is generated from Shape Net [6], a large-scale database of approximately 51K 3D models across 270 categories. [6] Angel X. Chang, Thomas Funkhouser, Leonidas Guibas, Pat Hanrahan, Qixing Huang, Zimo Li, Silvio Savarese, Manolis Savva, Shuran Song, Hao Su, Jianxiong Xiao, Li Yi, and Fisher Yu. Shape Net: An Information-Rich 3D Model Repository. arXiv:1512.03012, 2015. |
| Dataset Splits | No | We then compute the angular distance error on 10% of the models for each category held out as a test set. While a test set is mentioned, there's no explicit information about a validation split, nor are the train/validation/test splits provided in specific percentages or sample counts for all three. |
| Hardware Specification | No | The paper states 'We implemented our network in TensorFlow [1], and trained with the Adam optimizer... using synchronous training with 32 replicas.' but does not specify any particular hardware details such as GPU models, CPU types, or memory. |
| Software Dependencies | No | The paper states 'We implemented our network in TensorFlow [1]' but does not provide specific version numbers for TensorFlow or any other software dependencies. |
| Experiment Setup | Yes | We implemented our network in Tensor Flow [1], and trained with the Adam optimizer with a learning rate of 10 3, β1 = 0.9, β2 = 0.999, and a total batch size of 256. We use the following weights for the losses: (αcon, αpose, αsep, αobj) = (1, 0.2, 1.0, 1.0). We train the network for 200K steps using synchronous training with 32 replicas. |