AI IMAGES pretrained model

Here I am discovering pretrained neural models for images.

neural networks for pictures with artificial intelligence and Development|lower

NEURAL MODELS FOR IMAGES

SEGMENTATION

Object detection

SAM SEGMENT ANYTHING MODEL

1. From Meta. Trained on 11M images and 1.1B masks.

2. To install $ pip install git+https://github.com/facebookresearch/segment-anything.git

3. It can detect mask, objects.

4. If you get error -215:Assertion failed !_src.empty() in function 'cvtColor' // try instead of cv2.imread("file_path") to use:

from skimage import io

img = io.imread("file_path")

 

Image Aesthetics

MUSIQ multi-scale image quality

1. Trained on AVA dataset 250K images with descriptions.

2. github.com/google-research/google-research/tree/master/musiq

 

Improve resolution. Upscaling models.

1. stabilityai/stable-diffusion-x4-upscaler

 

 

Vila (VIsion Language Aesthetics) from Google

1. github.com/googleresearch/google-research/tree/master/vila // doesn't exist

2. Kaggle tf_hub example doesn't work

 

DEPTH ESTIMATION

1. Midas, Depth Anything, Zoo Depth