Update src/utils.py

src/utils.py

@@ -2,7 +2,6 @@
 
 import imp
 import numpy as np
-import cv2
 import torch
 import random
 from PIL import Image, ImageDraw, ImageFont
@@ -207,18 +206,6 @@ def load_512(image_path, left=0, right=0, top=0, bottom=0):
     image = np.array(Image.fromarray(image).resize((512, 512)))
     return image
 
-def get_canny(image_path):
-    image = load_512(
-        image_path
-    )
-    image = np.array(image)
-
-    # get canny image
-    image = cv2.Canny(image, 100, 200)
-    image = image[:, :, None]
-    image = np.concatenate([image, image, image], axis=2)
-    canny_image = Image.fromarray(image)
-    return canny_image
 
 
 def get_scribble(image_path, hed):
@@ -229,44 +216,7 @@ def get_scribble(image_path, hed):
 
     return image
 
-def get_cocoimages(prompt_path):
-    data_ls = []
-    with open(prompt_path) as f:
-        prompt_ls = json.load(f)
-    img_path = 'COCO2017-val/val2017'
-    for prompt in tqdm(prompt_ls):
-        caption = prompt['caption'].replace('/','_')
-        image_id = str(prompt['image_id'])
-        image_id = (12-len(image_id))*'0' + image_id+'.jpg'
-        image_path = os.path.join(img_path, image_id)
-        try:
-            image = get_canny(image_path)
-        except:
-            continue
-        curr_data = {'image':image, 'prompt':caption}
-        data_ls.append(curr_data)
-    return data_ls
-
-def get_cocoimages2(prompt_path):
-    """scribble condition
-    """
-    data_ls = []
-    with open(prompt_path) as f:
-        prompt_ls = json.load(f)
-    img_path = 'COCO2017-val/val2017'
-    hed = HEDdetector.from_pretrained('ControlNet/detector_weights/annotator', filename='network-bsds500.pth')
-    for prompt in tqdm(prompt_ls):
-        caption = prompt['caption'].replace('/','_')
-        image_id = str(prompt['image_id'])
-        image_id = (12-len(image_id))*'0' + image_id+'.jpg'
-        image_path = os.path.join(img_path, image_id)
-        try:
-            image = get_scribble(image_path,hed)
-        except:
-            continue
-        curr_data = {'image':image, 'prompt':caption}
-        data_ls.append(curr_data)
-    return data_ls
+
 
 def warpped_feature(sample, step):
     """