Transfer of View-manifold Learning to Similarity Perception of Novel Objects
Authors: Xingyu Lin, Hao Wang, Zhihao Li, Yimeng Zhang, Alan Yuille, Tai Sing Lee
ICLR 2017 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We develop a model of perceptual similarity judgment based on re-training a deep convolution neural network (DCNN) that learns to associate different views of each 3D object to capture the notion of object persistence and continuity in our visual experience. ... We demonstrate that multi-view association training with a relatively small set of objects directly affects similarity judgment across many classes of objects, including novel objects that the network has not seen before. ... We compare Ho G feature representation (Dalal & Triggs, 2005) and four deep learning networks: 1) OPnet, 2) Alex Net pre-trained on Image Net, 3) An Alex Net fine-tuned for classification on Shape Net data, denoted as Alex Net FT , 4) The joint embedding model by Li et al. (2015). ... We show the results for the instance retrieval task in Figure 2 and Table 1. |
| Researcher Affiliation | Academia | Xingyu Lin, Hao Wang Department of Computer Science Peking University Beijing, 100871, China {sean.linxingyu, hao.wang}@pku.edu.cn Zhihao Li, Yimeng Zhang Department of Computer Science Carnegie Mellon University Pittsburgh, PA 15213, USA {zhihaol, yimengzh}@andrew.cmu.edu Alan Yuille Department of Cognitive Science John Hopkins University Baltimore, MD 21218, USA alan.yuille@jhu.edu Tai Sing Lee Department of Computer Science Carnegie Mellon University Pittsburgh, PA 15213, USA tai@cnbc.cmu.edu |
| Pseudocode | No | The paper does not contain structured pseudocode or algorithm blocks. |
| Open Source Code | No | The paper does not provide concrete access to source code for the methodology described. |
| Open Datasets | Yes | We render multi-view images of individual objects from 7K 3D CAD models of objects in Shape Net (Chang et al., 2015). |
| Dataset Splits | Yes | For training, we sample 200 object models from 29 categories of Shape Net. 20 of these object models from each category are saved for cross validation. |
| Hardware Specification | No | The paper discusses training deep neural networks but does not provide specific hardware details such as GPU or CPU models used for the experiments. |
| Software Dependencies | No | We use Caffe (Jia et al., 2014) for training the networks. ... We use the rendering pipeline in Blender, an open source 3D graphics software, with a spotlight that is static relative to the camera. |
| Experiment Setup | Yes | Starting with a learning rate of 0.01, we decrease it by a factor of 10 every 8K iterations and with a momentum of 0.9. We stop the training at 20K iterations. Weight decay is set to 0.0005. We set the margin parameter M to 0.1 by cross validation. |