File size: 3,831 Bytes
71feab2 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 |
import gradio as gr
from transformers import AutoProcessor, AutoTokenizer, AutoImageProcessor, AutoModelForCausalLM, BlipForConditionalGeneration, Blip2ForConditionalGeneration, VisionEncoderDecoderModel
import torch
import open_clip
from huggingface_hub import hf_hub_download
git_processor_large_coco = AutoProcessor.from_pretrained("microsoft/git-large-coco")
git_model_large_coco = AutoModelForCausalLM.from_pretrained("microsoft/git-large-coco")
git_processor_large_textcaps = AutoProcessor.from_pretrained("microsoft/git-large-r-textcaps")
git_model_large_textcaps = AutoModelForCausalLM.from_pretrained("microsoft/git-large-r-textcaps")
blip_processor_large = AutoProcessor.from_pretrained("Salesforce/blip-image-captioning-large")
blip_model_large = BlipForConditionalGeneration.from_pretrained("Salesforce/blip-image-captioning-large")
blip2_processor_8_bit = AutoProcessor.from_pretrained("Salesforce/blip2-opt-6.7b")
blip2_model_8_bit = Blip2ForConditionalGeneration.from_pretrained("Salesforce/blip2-opt-6.7b", device_map="auto", load_in_8bit=True)
coca_model, _, coca_transform = open_clip.create_model_and_transforms(
model_name="coca_ViT-L-14",
pretrained="mscoco_finetuned_laion2B-s13B-b90k"
)
device = "cuda" if torch.cuda.is_available() else "cpu"
git_model_large_coco.to(device)
git_model_large_textcaps.to(device)
blip_model_large.to(device)
coca_model.to(device)
def generate_caption(processor, model, image, tokenizer=None, use_float_16=False):
inputs = processor(images=image, return_tensors="pt").to(device)
if use_float_16:
inputs = inputs.to(torch.float16)
generated_ids = model.generate(pixel_values=inputs.pixel_values, max_length=50)
if tokenizer is not None:
generated_caption = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
else:
generated_caption = processor.batch_decode(generated_ids, skip_special_tokens=True)[0]
return generated_caption
def generate_caption_coca(model, transform, image):
im = transform(image).unsqueeze(0).to(device)
with torch.no_grad(), torch.cuda.amp.autocast():
generated = model.generate(im, seq_len=20)
return open_clip.decode(generated[0].detach()).split("<end_of_text>")[0].replace("<start_of_text>", "")
def generate_captions(image):
caption_git_large_coco = generate_caption(git_processor_large_coco, git_model_large_coco, image)
caption_git_large_textcaps = generate_caption(git_processor_large_textcaps, git_model_large_textcaps, image)
caption_blip_large = generate_caption(blip_processor_large, blip_model_large, image)
caption_coca = generate_caption_coca(coca_model, coca_transform, image)
caption_blip2_8_bit = generate_caption(blip2_processor_8_bit, blip2_model_8_bit, image, use_float_16=True).strip()
return caption_git_large_coco, caption_git_large_textcaps, caption_blip_large, caption_coca, caption_blip2_8_bit
examples = [["Image1.jpg"], ["Image2.jpg"], ["Image3.jpg"]]
outputs = [gr.outputs.Textbox(label="Caption generated - 1"), gr.outputs.Textbox(label="Caption generated - 2"), gr.outputs.Textbox(label="Caption generated -3"), gr.outputs.Textbox(label="Caption generated - 4"), gr.outputs.Textbox(label="Caption generated - 5")]
title = "Interactive demo: comparing image captioning models"
description = "Image Caption Generator by Sravanth Kurmala"
article = "Assignment for Listed Inc"
interface = gr.Interface(fn=generate_captions,
inputs=gr.inputs.Image(type="pil"),
outputs=outputs,
examples=examples,
title=title,
description=description,
article=article,
enable_queue=True)
interface.launch(debug=True) |