modules/masking_module.py

#Script added by SPDraptor
import spaces
import copy
import numpy as np
import torch
from PIL import Image, ImageDraw
from transformers import  AutoProcessor, AutoTokenizer, AutoModelForCausalLM, BitsAndBytesConfig
from typing import Any
import supervision as sv
from sam2.build_sam import build_sam2, build_sam2_video_predictor
from sam2.sam2_image_predictor import SAM2ImagePredictor

device = torch.device('cuda')

model_id = 'microsoft/Florence-2-large'

models_dict = {
    'Florence_model':AutoModelForCausalLM.from_pretrained(model_id,
                                             trust_remote_code=True,
                                             attn_implementation="flash_attention_2",
                                             device_map=device).eval(),
    'Florence_processor':AutoProcessor.from_pretrained(model_id, trust_remote_code=True),
}

SAM_CHECKPOINT = "/home/user/app/sam2_hiera_large.pt"
SAM_CONFIG = "sam2_hiera_l.yaml"

@spaces.GPU
def load_sam_image_model(
    device: torch.device,
    config: str = SAM_CONFIG,
    checkpoint: str = SAM_CHECKPOINT
) -> SAM2ImagePredictor:
    model = build_sam2(config, checkpoint)
    return SAM2ImagePredictor(sam_model=model)


@spaces.GPU
def run_sam_inference(
    model: Any,
    image: Image,
    detections: sv.Detections
) -> sv.Detections:
    
    image = np.array(image.convert("RGB"))
    model.set_image(image)
    print(type(detections.xyxy),detections.xyxy)
    if detections.xyxy.size == 0:
      return {
                'code': 400,
                'data':'null',
                'message':'The AI couldn’t detect the object you want to mask.'
            }
      
    mask, score, _ = model.predict(box=detections.xyxy, multimask_output=False)

    # dirty fix; remove this later
    if len(mask.shape) == 4:
        mask = np.squeeze(mask)

    detections.mask = mask.astype(bool)
    return {
                'code': 200,
                'data':detections,
                'message':'The AI couldn’t detect the object you want to mask.'
            }
@spaces.GPU
def florence2(image,task_prompt, text_input=None):
    """
    Calling the Microsoft Florence2 model
    """
    model = models_dict['Florence_model']
    processor = models_dict['Florence_processor']
    # print(image)
    if text_input is None:
        prompt = task_prompt
    else:
        prompt = task_prompt + text_input

    input_florence = processor(text=prompt, images=image, return_tensors="pt").to("cuda")
    print(input_florence)
    generated_ids = model.generate(
        input_ids=input_florence["input_ids"],
        pixel_values=input_florence["pixel_values"],
        max_new_tokens=1024,
        early_stopping=False,
        do_sample=False,
        num_beams=3,
    )
    generated_text = processor.batch_decode(generated_ids,
                                            skip_special_tokens=False)[0]
    parsed_answer = processor.post_process_generation(
        generated_text,
        task=task_prompt,
        image_size=(image.width, image.height))

    return parsed_answer

def draw_MASK(image, prediction, fill_mask=False):
    """
    Draws segmentation masks with polygons on an image.

    Parameters:
    - image_path: Path to the image file.
    - prediction: Dictionary containing 'polygons' and 'labels' keys.
                  'polygons' is a list of lists, each containing vertices of a polygon.
                  'labels' is a list of labels corresponding to each polygon.
    - fill_mask: Boolean indicating whether to fill the polygons with color.
    """
    width=image.width
    height=image.height
    new_image = Image.new("RGB", (width, height), color="black")
    draw = ImageDraw.Draw(new_image)
    scale = 1

    for polygons, label in zip(prediction['polygons'], prediction['labels']):
        color = "white"
        fill_color = "white" if fill_mask else None

        for _polygon in polygons:
            _polygon = np.array(_polygon).reshape(-1, 2)
            if len(_polygon) < 3:
                print('Invalid polygon:', _polygon)
                continue

            _polygon = (_polygon * scale).reshape(-1).tolist()
            if fill_mask:
                draw.polygon(_polygon, outline=color, fill=fill_color)
            else:
                draw.polygon(_polygon, outline=color)
            draw.text((_polygon[0] + 8, _polygon[1] + 2), label, fill=color)

    return new_image

@spaces.GPU
def masking_process(image,obj):
    # task_prompt = '<REGION_TO_SEGMENTATION>'
    # # task_prompt = '<OPEN_VOCABULARY_DETECTION>'
    # print(type(task_prompt),type(obj))
    print('1')
    image = Image.fromarray(image).convert("RGB")

    # results = florence2(image,task_prompt, text_input=obj)
    # output_image = copy.deepcopy(image)
    # img=draw_MASK(output_image,
    #           results['<REGION_TO_SEGMENTATION>'],
    #           fill_mask=True)
    # mask=img.convert('1')
    task_prompt = '<OPEN_VOCABULARY_DETECTION>'

    # image = Image.open("/content/tiger.jpeg").convert("RGB")

    # obj = "Tiger"

    Florence_results = florence2(image,task_prompt, text_input=obj)
    print('2')
    SAM_IMAGE_MODEL = load_sam_image_model(device=device)
    print('3')
    detections = sv.Detections.from_lmm(
                lmm=sv.LMM.FLORENCE_2,
                result=Florence_results,
                resolution_wh=image.size
            )
    print('4')
    response = run_sam_inference(SAM_IMAGE_MODEL, image, detections)
    print('5')
    if response['code'] == 400:
        print("no object found")
        return "no object found"
    else:
        detections2=response['data']
        mask = Image.fromarray(detections2.mask[0])
        # response['data']=mask
        torch.cuda.empty_cache()
        return mask