Deep Learning Induced Background Matting
The goal of the project was to remove the background (mannequin) from the clothing display images, generated from the client’s shoot data. While many tools now provide background replacement functionality, they also yield artifacts at boundaries, particularly in areas where there are finer details. In contrast, traditional image matting methods provide much higher quality results, but do not run at high resolution, and frequently require manual input (Trimaps).
Hence, in order to achieve this ghost mannequin effect on the clothing, we used a technique based on background matting. In this approach, an additional frame of the background is captured and used in recovering the alpha matte and the foreground layer. The objective is to compute a high-quality alpha matte, preserving the edge details, while processing high-resolution images specified by the client.
About our Client
Client : Confidential
Deep Learning, PyTorch, Image Matting, Morphological Operations (OpenCV), Pandas, NumPy, Matplotlib, CUDA, Streamlit (frontend), Python (Uvicorn)