File size: 8,029 Bytes
ad44ad4 |
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 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 |
import os
import json
import cv2
import numpy as np
from pathlib import Path
import random
OBJECT_SIZE_TEMPLATES = {
"28": [
"There is a <>. Adjust the dimensions of <> to make it 1.2 times its original size.",
"There is a <>. Enlarge the <> such that it becomes 1.2 times its starting size.",
"There is a <>. Ensure the <> grows to precisely 1.2 times its original size.",
"There is a <>. Increase the size of <> to 1.2 times its original size.",
"There is a <>. Resize the <> to ensure it is 1.2 times its original dimensions.",
"There is a <>. Scale the <> up so that it reaches 1.2 times its initial dimensions.",
"There is a <>. Transform the <> by setting its size to 1.2 times its original value.",
],
"29": [
"There is a <>. Adjust the height of <> to be 20cm taller.",
"There is a <>. Enlarge the <> in height by 20cm.",
"There is a <>. Increase the height of <> by 20cm.",
"There is a <>. Raise <>'s height by 20cm",
],
"30": [
"There is a <>. Adjust the length of <> to be 50cm longer.",
"There is a <>. Enlarge the <> in length by 50cm.",
"There is a <>. Extend <>'s length by 50cm",
"There is a <>. Increase the length of <> by 50cm.",
],
"31": [
"There is a <>. Adjust the width of <> to be 40cm wider.",
"There is a <>. Enlarge the <> in width by 40cm.",
"There is a <>. Increase the width of <> by 40cm.",
"There is a <>. Widen <>'s width by 40cm",
],
}
OPERATIONS = {
"28": (1.3, 1.3),
"29": (1.0, 1.3),
"30": (1.3, 1.0),
"31": (1.3, 1.0),
}
def is_mask_single_component(mask_path):
mask = cv2.imread(mask_path, cv2.IMREAD_GRAYSCALE)
if mask is None:
return False
_, binary = cv2.threshold(mask, 127, 255, cv2.THRESH_BINARY)
num_labels, _ = cv2.connectedComponents(binary)
return (num_labels - 1) == 1
def resize_and_paste_back_with_repair_mask(
image_path, mask_path, scale_width=1.0, scale_height=1.0, output_image_path="output.png"
):
image = cv2.imread(image_path)
mask = cv2.imread(mask_path, cv2.IMREAD_GRAYSCALE)
if image is None or mask is None:
raise FileNotFoundError("Image or mask not found!")
h, w = image.shape[:2]
assert mask.shape == (h, w), "Mask and image size mismatch!"
coords = cv2.findNonZero(mask)
if coords is None:
raise ValueError("No object in mask!")
x, y, obj_w, obj_h = cv2.boundingRect(coords)
center_x = x + obj_w // 2
center_y = y + obj_h // 2
rgba = np.dstack([image, mask])
obj_rgba = rgba[y:y+obj_h, x:x+obj_w]
new_w = max(1, int(obj_w * scale_width))
new_h = max(1, int(obj_h * scale_height))
resized_obj = cv2.resize(obj_rgba, (new_w, new_h), interpolation=cv2.INTER_AREA)
new_x = center_x - new_w // 2
new_y = center_y - new_h // 2
dst_x_start = max(0, new_x)
dst_y_start = max(0, new_y)
dst_x_end = min(w, new_x + new_w)
dst_y_end = min(h, new_y + new_h)
src_x_start = max(0, -new_x)
src_y_start = max(0, -new_y)
src_x_end = min(new_w, w - dst_x_start)
src_y_end = min(new_h, h - dst_y_start)
background = image.copy()
background[mask > 0] = [255, 255, 255]
output_img = background.copy()
if src_x_end > src_x_start and src_y_end > src_y_start:
obj_part = resized_obj[src_y_start:src_y_end, src_x_start:src_x_end]
alpha = obj_part[:, :, 3].astype(np.float32) / 255.0
bg_part = output_img[dst_y_start:dst_y_end, dst_x_start:dst_x_end]
fg_part = obj_part[:, :, :3]
# 防止任何极端情况下 shape 不一致
hh = min(fg_part.shape[0], bg_part.shape[0], alpha.shape[0])
ww = min(fg_part.shape[1], bg_part.shape[1], alpha.shape[1])
fg_part = fg_part[:hh, :ww]
bg_part = bg_part[:hh, :ww]
alpha = alpha[:hh, :ww]
blended = fg_part * alpha[..., None] + bg_part * (1 - alpha[..., None])
output_img[dst_y_start:dst_y_start+hh, dst_x_start:dst_x_start+ww] = blended.astype(np.uint8)
os.makedirs(os.path.dirname(output_image_path), exist_ok=True)
cv2.imwrite(output_image_path, output_img)
return output_img
def load_object_class_mapping(genspace_json_path):
with open(genspace_json_path, "r", encoding="utf-8") as f:
data = json.load(f)
return {s["sample_id"]: s["object_class"] for s in data["samples"]}
def main():
MASK_DIR = "/mnt/prev_nas/qhy_1/datasets/flux_gen_images_masks"
IMAGE_DIR = "/mnt/prev_nas/qhy_1/datasets/flux_gen_images"
OUTPUT_DIR = "/mnt/prev_nas/qhy_1/datasets/flux_gen_images_size_change"
GENSPACE_JSON = "/mnt/prev_nas/qhy_1/datasets/unedit_image_prompts/genspace_prompts_vlm.json"
OUTPUT_JSONL_PATH = os.path.join(OUTPUT_DIR, "size_edit_annotations.jsonl")
os.makedirs(OUTPUT_DIR, exist_ok=True)
obj_class_map = load_object_class_mapping(GENSPACE_JSON)
mask_files = sorted([f for f in os.listdir(MASK_DIR) if f.endswith(".png")])
print(f"Found {len(mask_files)} mask files.")
num_imgs_generated = 0
num_lines_written = 0
num_skipped_multi = 0
num_missing_img = 0
with open(OUTPUT_JSONL_PATH, "w", encoding="utf-8") as f_out:
for mask_file in mask_files:
mask_path = os.path.join(MASK_DIR, mask_file)
if not is_mask_single_component(mask_path):
num_skipped_multi += 1
print(f"💥 Skipping multi-component mask: {mask_file}")
continue
stem = Path(mask_file).stem # e.g. allocentric_002
object_class = obj_class_map.get(stem, None)
if object_class is None:
# 没有对应 class 的就跳过
print(f"💥 No class for {stem}")
continue
image_path = os.path.join(IMAGE_DIR, mask_file) # same name
if not os.path.exists(image_path):
num_missing_img += 1
print(f"💥 Missing image for {stem}")
continue
for task_id, (scale_w, scale_h) in OPERATIONS.items():
output_name = f"{stem}_{task_id}.png"
output_path = os.path.join(OUTPUT_DIR, output_name)
# 1) 图片已存在:不再生成
if not os.path.exists(output_path):
try:
resize_and_paste_back_with_repair_mask(
image_path=image_path,
mask_path=mask_path,
scale_width=scale_w,
scale_height=scale_h,
output_image_path=output_path,
)
num_imgs_generated += 1
except Exception as e:
print(f"💥 Error generating {output_name}: {e}")
continue
else:
print(f"✅ Already exists: {output_name}")
# 2) 无论是否生成,jsonl 都写一条(只写需要内容)
template = random.choice(OBJECT_SIZE_TEMPLATES[task_id])
instruction = template.replace("<>", f"<{object_class}>")
item = {
"task_type": "edit",
"instruction": instruction,
"input_images": [image_path],
"output_image": output_path,
"object_class": object_class,
"task_id": int(task_id),
}
f_out.write(json.dumps(item, ensure_ascii=False) + "\n")
num_lines_written += 1
print("\nDone.")
print("images generated:", num_imgs_generated)
print("jsonl lines written:", num_lines_written)
print("skipped (multi-components):", num_skipped_multi)
print("missing images:", num_missing_img)
print("jsonl saved to:", OUTPUT_JSONL_PATH)
if __name__ == "__main__":
main()
|