The image resizer A.I. Gigapixel, from Topaz Labs, received a new update, featuring additional options for image enhancement, 3 sets of new Artificial Intelligence models and other features.
With the latest update to A.I.Gigapixel the program, which Topaz Labs presents as “the first and only intelligent image resizer for your computer” just became even more powerful. The company not only refined the A.I. in the existing models used to resize images, for more accurate results, also introduced additional options for image enhancement, 3 sets of new Artificial Intelligence models to resize high-quality input images or even images with a lot of noise.
One of the promised features is introduced in the new update: A.I. Gigapixel now offers CPU processing, so users with humble GPU can use the program. As mentioned before here at ProVideo Coalition, the new stand-alone software from Topaz Labs needs a lot of computer power to work, and depends heavily on the GPU to process images. In fact, A.I. Gigapixel performs millions of calculations per pixel in your image, resulting in trillions of computations for your image to ensure it enhances all the right details. Applying all these calculations on a full-size RAW image can take time to complete, but the results speak for themselves.
According to Topaz Labs the A.I. Gigapixel V1.1 update, results from the substantial improvements made on the deep learning models behind the scenes by its research scientist, Dr. Acharjee. Those enhancements are reflected as replacing the “Enhancement” checkbox with a multi-level “Reduce Noise and Blur” options that offers three 3 sets of neural networks tailored for images of different quality, giving users a more precise control over the final results.
Although the difference is, according to Topaz Labs, very subtle for most photos, for which the default setting (“Moderate”) will work well, users may want to try different “Reduction Noise and Blur” settings when aiming for the highest quality result. The options now available allow users to select one of the following neural networks:
- “None”: This neural network was trained with clean images. If your original photo is a well exposed RAW or is otherwise free of visible artifacts, this option will best preserve and create fine detail.
- “Moderate”: This default option will suppress a moderate amount of image noise and JPEG compression artifacts and apply some sharpening. This setting is good for most consumer-level standard cameras and phone photos.
- “Strong”: This option is for noisy, highly compressed, or otherwise artifacted images. This neural network applies the heaviest level of sharpening and noise reduction. Unfortunately, if too much information is missing or obscured by noise, A.I. cannot synthesize details properly, and sometimes produce strange structures on details like faces. Alas, A.I. is not quite that magical when it comes to recreating faces yet. Therefore, do not use this setting on clean photos, that don’t have much noise.
The update is free for owners of the A.I. Gigapixel software. As always, adds Topaz Labs, “updates to our products are free for products you already own, so if you already own Gigapixel you get all these new features for free!”. Meanwhile the company continues to accept feedback from users and even reply to some of the suggestions with a “we will investigate” as Albert Yang, Topaz Labs CEO, noted on the comment from a user requesting that A.I. Gigapixel works as a plugin.
The company also continues to challenge photographers to try the software, saying that “since Gigapixel will not overwrite the original images, you can experiment with each setting to your heart’s content.” One important information from the online conversation is the note that the new update makes image processing slower for image scales under 220%. The note adds that “Dr. Acharjee has made substantial progress on the neural network architecture for a considerable increase in image quality. According to feedback from the general community, speed takes a backseat to quality results. For scales greater than 220%, different optimizations were implemented that will not result in longer processing times.”