Captioning is an img2txt model that uses the BLIP. Exports captions of images.
If there is no 'Checkpoints' folder, the script will automatically create the folder and download the model file, you can do this manually if you want.
Download the fine-tuned checkpoint and copy into 'checkpoints' folder (create if does not exists)
datasets\0.jpg, a piece of cheese with figs and a piece of cheese
datasets\1002.jpg, a close up of a yellow flower with a green background
datasets\1005.jpg, a planter filled with lots of colorful flowers
datasets\1008.jpg, a teacher standing in front of a classroom full of children
datasets\1011.jpg, a tortoise on a white background with a white background
datasets\1014.jpg, a glass of wine sitting on top of a table
datasets\1017.jpg, a close up of a plant with pink flowers
datasets\102.jpg, a platter of different types of sushi
datasets\1020.jpg, a frog sitting on top of a bamboo stick
datasets\1023.jpg, a revolver on a white background
datasets\1026.jpg, a woman holding a small white dog in her arms
datasets\1029.jpg, a woman in a business suit standing in front of a building
datasets\1032.jpg, sliced cucumber on a white background
datasets\1035.jpg, a woman in glasses and a pair of boxing gloves
datasets\1038.jpg, a pile of sliced potatoes on a white surface
datasets\1041.jpg, two glasses of orange juice on a wooden table
datasets\1044.jpg, a woman sitting on the floor in front of a door
usage: inference.py [-h] [-i INPUT] [-b BATCH] [-p PATHS] [-g GPU_ID]
Image caption CLI
optional arguments:
-h, --help show this help message and exit
-i INPUT, --input INPUT Input directoryt path, such as ./images
-b BATCH, --batch BATCH Batch size
-p PATHS, --paths PATHS A any.txt files contains all image paths.
-g GPU_ID, --gpu-id GPU_ID gpu device to use (default=0) can be 0,1,2 for multi-gpupython inference.py -i /path/images/folder --batch 8 --gpu 0Pull requests are welcome. For major changes, please open an issue first to discuss what you would like to change.
Please make sure to update tests as appropriate.