CoSE: Compositional Stroke Embeddings
Authors: Emre Aksan, Thomas Deselaers, Andrea Tagliasacchi, Otmar Hilliges
NeurIPS 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We demonstrate qualitatively and quantitatively that our proposed approach is able to model the appearance of individual strokes, as well as the compositional structure of larger diagram drawings. Our approach is suitable for interactive use cases such as auto-completing diagrams. We make code and models publicly available at https://eth-ait.github.io/cose. |
| Researcher Affiliation | Collaboration | Emre Aksan ETH Zurich eaksan@inf.ethz.ch Thomas Deselaers Apple Switzerland deselaers@gmail.com Andrea Tagliasacchi Google Research atagliasacchi@google.com Otmar Hilliges ETH Zurich otmar.hilliges@inf.ethz.ch Work done while at Google. Unrelated to affiliation with Apple |
| Pseudocode | No | The paper describes its architecture and methods verbally and with diagrams (e.g., Figure 2), but no formal pseudocode blocks or algorithms are explicitly provided or labeled. |
| Open Source Code | Yes | We make code and models publicly available at https://eth-ait.github.io/cose. |
| Open Datasets | Yes | We evaluate our model on the recently released Di Di dataset [11]. In this paper we focus on the predictive setting in which an existing (partial) drawing is extended by adding more shapes or by connecting already drawn ones... We experimentally show that the architecture can model complex diagrams and flow-charts from the Di Di dataset, free-form sketches from Quick Draw and handwritten text from the IAM-On DB datasets [13]. |
| Dataset Splits | No | The paper mentions 'training set' ('We train by maximizing the log-likelihood of the network parameters θ on the training set.') and 'test data' but does not explicitly state a separate validation set or its split percentage/counts. |
| Hardware Specification | No | No specific hardware details (e.g., GPU/CPU models, memory, cloud instances, or machine specifications) are mentioned in the paper for the experiments. |
| Software Dependencies | No | The paper mentions software components and frameworks like 'transformers' and 'LSTMs', but does not provide specific version numbers for any software dependencies or libraries (e.g., PyTorch version, Python version, specific library versions). |
| Experiment Setup | No | The paper mentions training details like data augmentation ('random rotation and re-scaling of the entire drawing (see supplementary for details)'), the use of 'mixture densities with M Gaussians', and discusses the 'number of GMM components' in ablation studies (Fig. 9). However, specific hyperparameters such as learning rate, batch size, or number of epochs are not explicitly stated in the main text, with some details deferred to supplementary material. |