Synthetic Faces with Landmark Annotations and Captions

Face Synthetics Spiga Captioned is a copy of the Microsoft FaceSynthetics dataset enhanced with SPIGA-calculated facial landmark annotations and BLIP-generated text captions. The dataset, created by multimodalart and last updated in March 2023, is designed for multimodal tasks involving synthetic facial imagery.

Use Cases

• Train a facial landmark detection model using the SPIGA-calculated landmark annotations on synthetic faces.
• Develop an image captioning model by training on BLIP-generated captions paired with synthetic face images.
• Fine-tune a generative model for face synthesis using the landmark annotations as conditioning signals.
• Build a multimodal retrieval system linking synthetic facial images to their corresponding text captions.

Strengths

• Dataset includes multiple annotation types: synthetic images, facial landmarks, and text captions.
• Based on the established Microsoft FaceSynthetics dataset, providing a known synthetic foundation.
• Annotations are algorithmically generated using SPIGA and BLIP models for consistency.

Limitations

• All images are synthetic, which may limit applicability to models intended for real-world facial data.
• Annotations are generated by models (SPIGA, BLIP) and may contain inherent errors or biases from those tools.
• Dataset size, row count, and specific file formats are not provided in the available information.

Provenance

Source

Derived from the Microsoft FaceSynthetics dataset.

Collection Method

Original synthetic faces augmented with SPIGA-calculated facial landmarks and BLIP-generated captions.

Freshness

Last updated on 2023-03-23.