Cousin Network Guided Sketch Recognition via Latent Attribute Warehouse
Authors: Kaihao Zhang, Wenhan Luo, Lin Ma, Hongdong Li9203-9210
AAAI 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments conducted on the TU-Berlin dataset show that the proposed model is able to efficiently distill knowledge from natural images and achieves superior performance than the current state of the art. |
| Researcher Affiliation | Academia | Kaihao Zhang,1,3 Wenhan Luo,2 Lin Ma,2 Hongdong Li1,3 1Australian National University 3Australian Centre for Robotic Vision {kaihao.zhang, hongdong.li}@anu.edu.au quad {whluo.china, forest.linma}@gmail.com |
| Pseudocode | No | The paper does not contain any structured pseudocode or algorithm blocks. |
| Open Source Code | No | The paper does not provide explicit statements or links for the open-sourcing of the code for their methodology. |
| Open Datasets | Yes | TU-Berlin Dataset. This dataset is proposed in (Eitz, Hays, and Alexa 2012) for sketch recognition, which contains 250 classes of objects. Each class has 80 sketches. TU-Berlin-extend Dataset. The TU-Berlin dataset is extended by adding real images (Zhang et al. 2016). |
| Dataset Splits | No | The paper describes training and testing splits: 'For each protocol, a number of t sketch instances in each category is used for training and the rest sketches in the category are used for testing. Values of t are 16, 24, 32, 40, 48, 56, 64 and 72 for these protocols.' However, it does not explicitly define a separate validation split. |
| Hardware Specification | No | The paper does not provide specific hardware details (e.g., GPU/CPU models, memory) used for running its experiments. |
| Software Dependencies | No | The paper does not specify any software dependencies with version numbers (e.g., Python, PyTorch, TensorFlow versions). |
| Experiment Setup | Yes | In the training stage, the input sketch and images are both resized to 256 256. We crop 224 224 3 patches from the resized images and flip them horizontally at random. All weights are initialized as a Gaussian distribution (mean=0 and standard deviation = 0.02). Momentum is set at 0.9. α and β are set to 0.3. The whole training procedure is as follows. Firstly, The CNG-SCN model is trained without warehouse module. Then we fix the above pre-trained model and train the warehouse individually. After that, the CNG-SCN model is combined with warehouse module and fine-turned to obtain a new model. We combine the two kinds of CNGSCN models to make final prediction. |