CG-GAN: An Interactive Evolutionary GAN-Based Approach for Facial Composite Generation
Authors: Nicola Zaltron, Luisa Zurlo, Sebastian Risi2544-2551
AAAI 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | A preliminary testing phase was performed, which aimed at understanding how non-experts interact with the system. The study involved a diverse group of users and allowed us to fine-tune the interface; some results are shown in Figure 9. For the final user test, participants were divided in two groups: 13 constructors created composites and 55 evaluators evaluated them. |
| Researcher Affiliation | Academia | Nicola Zaltron IT University of Copenhagen Copenhagen, Denmark nicolazaltra@gmail.com Luisa Zurlo IT University of Copenhagen Copenhagen, Denmark zurlo.luisa@gmail.com Sebastian Risi IT University of Copenhagen Copenhagen, Denmark sebr@itu.dk |
| Pseudocode | No | The paper describes the process and provides equations for mutations but does not include any pseudocode or algorithm blocks. |
| Open Source Code | Yes | The questionnaire, its results, a demo and the code for the experiments in this paper can be found at: https://github.com/Luisa Zurlo/CG-GAN. |
| Open Datasets | Yes | In this paper, we employ a pg-GAN1 pre-trained on the Celeb Faces Attributes Dataset (Liu and others 2015), which contains 200,000 images of celebrity faces, annotated with 40 binary attributes. |
| Dataset Splits | No | The paper mentions using the Celeb Faces Attributes Dataset but does not specify how it was split into training, validation, or test sets for the GAN training, nor does it provide explicit splits for the user study data. |
| Hardware Specification | No | The paper does not provide specific details about the hardware used for running the experiments (e.g., GPU models, CPU types). |
| Software Dependencies | No | The paper mentions using a 'pg-GAN' and 'TL-GAN' approach and custom models, but it does not specify versions for any programming languages, libraries, or frameworks (e.g., Python version, PyTorch/TensorFlow version). |
| Experiment Setup | Yes | Strict rules were set on the timing: every user had 15 minutes to try the software before starting. The sessions lasted up to 30 minutes and data was collected every 10 minutes, to compare different phases of the process. ... Three mutation types are available: random changes, one unlocked feature and every unlocked features. ... The amount depends on a Gaussian distribution with μ and σ defined as: μ = 20 desired changes amount where F is set to 1 in case of one unlocked feature and F = min(max(1, (0.8 #unlocked features)), 8), in case of every unlocked feature. The specific values were finetuned through prior experimentation. |