Cutting out big product photos: mask resizing cost 1.5 of 255, the model cost 54

A young entrepreneur faced the challenge of cutting out product photos from high-resolution images within a tight timeframe. The task required resizing masks and resizing images, both of which had an impact on the overall project cost.

The entrepreneur tested three different models offered by a browser background remover tool, ImgIng. The three tiers available were fast, pro, and heavy. Each model had its own resolution, with the fast tier processing at 1024x1024, the pro tier at 1024x1024, and the heavy tier at 2048x2048. These resolutions significantly altered the masks produced by the model, ultimately affecting the overall quality of the final product images.

When the heavy model was tested on a 2048x2048 photo, the mask quality dropped from 0.975 to 0.223. This led to the conclusion that the model was trained for a specific resolution and that resizing the input photo to match the model's native resolution would result in better mask quality. The entrepreneur discovered that resizing the mask was a more cost-effective solution than attempting to upscale the original image.

The entrepreneur calculated the cost of resizing the mask to be around 1 to 1.5 out of 255, while the cost of the model remained constant at 54. Thus, resizing the mask proved to be a cheaper alternative compared to using a more powerful model. The entrepreneur decided to categorize the photos based on their type and select the appropriate model to use for each, ultimately reducing the overall cost of the project.