PyCinemetricsV2: Interactive computational film software based on transformers and PySide6
Year: 2025
Type: software-paper
Abstract: Based on feedback from film scholars, we developed PyCinemetricsV2, an upgraded film analytics software. Built on pre-trained AIGC models and PySide6, it offers features such as removing false positives and adding false negatives after TransNetV2 Shot-Boundary Detection (SBD), as well as generating configurable shot mosaic. Using the multimodal Git-base model, it provides descriptions of shot frames and enables open-domain object recognition. Using Faster-Whisper, it delivers more accurate dialog recognition, while PaddleOCR is employed to identify film metadata and intertitles. Based on InsightFace, facial recognition is implemented for the shot-first frames or for fine-grained frames, generating a list of credits. The case study demonstrates its accuracy, efficiency, and improved user experience in film analytics.
Source: SoftwareX
Authors: Chunfang Li, Yalv Fan, Yushi Shen, Kun Wang, Yuhe Hu, Fei Zhang, Yuchen Pei, Tongtong Zheng, Zhuoqi Shi
Institution: Communication University of China
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: gold
Estimated APC paid: $1,200
Awards: G3452794769, G6221437324, G7771819905
OpenAlex ID: W4413290089