Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in [1].

Learning to Infer Graphics Programs from Hand-Drawn Images

Authors: Kevin Ellis, Daniel Ritchie, Armando Solar-Lezama, Josh Tenenbaum

NeurIPS 2018 | Venue PDF | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We trained our network by sampling specs S and target images I for randomly generated scenes1 and maximizing Pθ[S|I], the likelihood of S given I, with respect to model parameters θ, by gradient ascent. We trained on 105 scenes... Experiment 1: Figure 4. To evaluate which components of the model are necessary to parse complicated scenes, we compared the neural network with SMC against the neural network by itself... Experiment 2: Figures 5 7. We evaluated, but did not train, our system on 100 real hand-drawn figures... Experiment 3: Table 5; Figure 8; Supplement Section 4. We compare synthesis times for our learned search policy with 4 alternatives...
Researcher Affiliation Academia Kevin Ellis MIT EMAIL Daniel Ritchie Brown University EMAIL Armando Solar-Lezama MIT EMAIL Joshua B. Tenenbaum MIT EMAIL
Pseudocode No The paper describes algorithms and presents a grammar (Table 2) but does not include structured pseudocode or clearly labeled algorithm blocks.
Open Source Code Yes The code and data are available at https://github.com/ellisk42/Tik Z.
Open Datasets Yes We trained our network by sampling specs S and target images I for randomly generated scenes1... We trained on 105 scenes... We evaluated, but did not train, our system on 100 real hand-drawn figures... The code and data are available at https://github.com/ellisk42/Tik Z.
Dataset Splits No The paper mentions training on '10^5 scenes' and evaluating on '100 real hand-drawn figures' but does not specify explicit training, validation, or test dataset splits or percentages.
Hardware Specification Yes We trained on 105 scenes, which takes a day on an Nvidia Titan X GPU.
Software Dependencies No The paper mentions using tools like 'Sketch tool [1]' but does not provide specific version numbers for any software, libraries, or programming languages (e.g., Python, PyTorch) used in the experiments.
Experiment Setup No The paper states 'Supplement Section 1 gives the full details of the architecture and training of this network' but does not provide specific hyperparameter values, optimizer settings, or other detailed training configurations in the main text.