v0.0.17

• Detection post-alignment now validates projected coordinates against base image size to prevent failures due to invalid coordinates (#100)

Package

https://pypi.org/project/retina-face/0.0.17/

v0.0.16

🎉 We're thrilled to announce the release of RetinaFace version 0.0.16! This update brings exciting improvements and fixes, making RetinaFace even more powerful and compatible. Here are the highlights:

✨ Enhanced Compatibility: RetinaFace is now fully compatible with TensorFlow versions 2.16 and later, ensuring smooth performance and integration with the latest TensorFlow environments. (#99)

🛠️ Alignment Improvement: In previous versions, alignment could encounter issues when encountering a zero-length line, resulting in alignment troubles due to the angle becoming undefined. With this update, such scenarios are handled more effectively, preventing alignment hiccups. (#95)

Upgrade now to experience these enhancements and more in RetinaFace 0.0.16! Happy detecting! 🚀

v0.0.15

🚀 RetinaFace Project Release Notes - Version 0.0.15

🔧 Improvements and Fixes:

• Resolved issue with TensorFlow's new versions where setting 'VALID' to padding threw exceptions; now expects lowercase 'valid'. (#91)
• Reordered process to align first and detect second, reducing meaningless black pixels. (#80, #85)
• Added package version to RetinaFace model, accessible upon import. (#81)
• Fixed bug related to loading images from the web. (#83)
• Enhanced logic for expanding facial area. (#87)

Package

https://pypi.org/project/retina-face/0.0.15/

v0.0.14

🛠️ GitHub Actions & Code Quality Improvements: GitHub Actions have been enabled for streamlined workflows.
Unit tests have been modularized, improving maintainability. Linting is now part of the development process to ensure code consistency. Type hinting has been added for input arguments and return types, enhancing code readability and developer experience. (PR-73)

📜 Logger Class Implementation: A dedicated logger class has been introduced, replacing the use of built-in print statements. This improves logging capabilities and facilitates better debugging. (PR-73)

🖼️ Extended Input Support: Support for URLs and base64-encoded images has been added to accommodate a broader range of input sources. This feature enhances the versatility of the software. (PR-73)

🔍 Optional Re-alignment for Single Face Images: For images with only one face, there's now an option to align, detect, and align again. This addresses the issue of black pixels after alignment, resulting in improved image quality and aesthetics. (PR-75)

🔍 Face Extraction Enhancement: Added an "expand_face_area" argument to the extract_faces() function (Line 206). Now, the function is capable of returning a larger area around the face by the specified expansion factor. This provides more flexibility in face extraction. (PR-49, PR-75)

🔄 Updated Numpy Types: Deprecated numpy int and float types have been replaced with built-in Python's int and float types. This enhances compatibility and ensures adherence to current best practices. (PR-52)

🖥️ Limited VRAM Reservation: The amount of reserved VRAM has been limited, allowing other scripts to run concurrently on the same GPU. This optimization improves resource utilization and system efficiency. (PR-65)

Package

• https://pypi.org/project/retina-face/0.0.14/