Spaces:
Paused
Paused
Update app.py
Browse files
app.py
CHANGED
@@ -1,3 +1,5 @@
|
|
|
|
|
|
1 |
import os
|
2 |
import sys
|
3 |
import cv2
|
@@ -6,48 +8,162 @@ import torch
|
|
6 |
import gradio as gr
|
7 |
from PIL import Image, ImageFilter, ImageDraw
|
8 |
|
9 |
-
|
10 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
11 |
|
|
|
|
|
|
|
|
|
|
|
|
|
12 |
|
13 |
-
|
14 |
-
|
15 |
-
import math
|
16 |
-
from utils.utils import get_bbox_from_mask, expand_bbox, pad_to_square, box2squre, crop_back, expand_image_mask
|
17 |
|
18 |
-
import os,sys
|
19 |
-
os.system("python -m pip install -e segment_anything")
|
20 |
-
os.system("python -m pip install -e GroundingDINO")
|
21 |
sys.path.append(os.path.join(os.getcwd(), "GroundingDINO"))
|
22 |
sys.path.append(os.path.join(os.getcwd(), "segment_anything"))
|
23 |
-
os.system("wget https://huggingface.co/ShilongLiu/GroundingDINO/resolve/main/groundingdino_swinb_cogcoor.pth")
|
24 |
-
os.system("wget https://huggingface.co/spaces/mrtlive/segment-anything-model/resolve/main/sam_vit_h_4b8939.pth")
|
25 |
-
import torchvision
|
26 |
-
from GroundingDINO.groundingdino.util.inference import load_model
|
27 |
-
from segment_anything import build_sam, SamPredictor
|
28 |
-
import spaces
|
29 |
-
import GroundingDINO.groundingdino.datasets.transforms as T
|
30 |
-
from GroundingDINO.groundingdino.util.utils import clean_state_dict, get_phrases_from_posmap
|
31 |
|
|
|
|
|
|
|
|
|
|
|
|
|
32 |
|
|
|
|
|
|
|
|
|
|
|
|
|
33 |
|
34 |
# GroundingDINO config and checkpoint
|
35 |
GROUNDING_DINO_CONFIG_PATH = "./GroundingDINO_SwinB.cfg.py"
|
36 |
-
GROUNDING_DINO_CHECKPOINT_PATH = "
|
37 |
|
38 |
# Segment-Anything checkpoint
|
39 |
SAM_ENCODER_VERSION = "vit_h"
|
40 |
-
SAM_CHECKPOINT_PATH = "
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
41 |
|
42 |
-
#
|
43 |
-
groundingdino_model = load_model(model_config_path=GROUNDING_DINO_CONFIG_PATH, model_checkpoint_path=GROUNDING_DINO_CHECKPOINT_PATH, device="cuda")
|
44 |
-
# Building SAM Model and SAM Predictor
|
45 |
sam = build_sam(checkpoint=SAM_CHECKPOINT_PATH)
|
46 |
sam.to(device="cuda")
|
47 |
sam_predictor = SamPredictor(sam)
|
48 |
|
49 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
50 |
|
|
|
|
|
|
|
|
|
51 |
transform = T.Compose(
|
52 |
[
|
53 |
T.RandomResize([800], max_size=1333),
|
@@ -60,80 +176,54 @@ def transform_image(image_pil):
|
|
60 |
|
61 |
|
62 |
def get_grounding_output(model, image, caption, box_threshold=0.25, text_threshold=0.25, with_logits=True):
|
63 |
-
caption = caption.lower()
|
64 |
-
caption = caption.strip()
|
65 |
if not caption.endswith("."):
|
66 |
caption = caption + "."
|
67 |
-
|
68 |
with torch.no_grad():
|
69 |
outputs = model(image[None], captions=[caption])
|
70 |
logits = outputs["pred_logits"].cpu().sigmoid()[0] # (nq, 256)
|
71 |
-
boxes = outputs["pred_boxes"].cpu()[0]
|
72 |
-
logits.shape[0]
|
73 |
|
74 |
# filter output
|
75 |
-
|
76 |
-
|
77 |
-
|
78 |
-
logits_filt = logits_filt[filt_mask] # num_filt, 256
|
79 |
-
boxes_filt = boxes_filt[filt_mask] # num_filt, 4
|
80 |
-
logits_filt.shape[0]
|
81 |
|
82 |
# get phrase
|
83 |
tokenlizer = model.tokenizer
|
84 |
tokenized = tokenlizer(caption)
|
85 |
-
|
86 |
-
pred_phrases = []
|
87 |
-
scores = []
|
88 |
for logit, box in zip(logits_filt, boxes_filt):
|
89 |
-
pred_phrase = get_phrases_from_posmap(
|
90 |
-
|
91 |
-
if with_logits:
|
92 |
-
pred_phrases.append(
|
93 |
-
pred_phrase + f"({str(logit.max().item())[:4]})")
|
94 |
-
else:
|
95 |
-
pred_phrases.append(pred_phrase)
|
96 |
scores.append(logit.max().item())
|
97 |
-
|
98 |
return boxes_filt, torch.Tensor(scores), pred_phrases
|
99 |
|
100 |
|
101 |
def get_mask(image, label):
|
102 |
global groundingdino_model, sam_predictor
|
103 |
-
|
104 |
-
|
105 |
image_pil = image.convert("RGB")
|
106 |
transformed_image = transform_image(image_pil)
|
107 |
|
108 |
-
|
109 |
boxes_filt, scores, pred_phrases = get_grounding_output(
|
110 |
groundingdino_model, transformed_image, label
|
111 |
)
|
112 |
|
113 |
-
|
114 |
-
|
115 |
-
# process boxes
|
116 |
-
H, W = size[1], size[0]
|
117 |
for i in range(boxes_filt.size(0)):
|
118 |
boxes_filt[i] = boxes_filt[i] * torch.Tensor([W, H, W, H])
|
119 |
boxes_filt[i][:2] -= boxes_filt[i][2:] / 2
|
120 |
boxes_filt[i][2:] += boxes_filt[i][:2]
|
121 |
-
|
122 |
boxes_filt = boxes_filt.cpu()
|
123 |
|
124 |
-
|
125 |
-
|
126 |
-
nms_idx = torchvision.ops.nms(
|
127 |
-
boxes_filt, scores, 0.8).numpy().tolist()
|
128 |
boxes_filt = boxes_filt[nms_idx]
|
129 |
-
pred_phrases = [pred_phrases[idx] for idx in nms_idx]
|
130 |
-
|
131 |
-
|
132 |
-
image = np.array(image_pil)
|
133 |
-
sam_predictor.set_image(image)
|
134 |
|
|
|
|
|
135 |
transformed_boxes = sam_predictor.transform.apply_boxes_torch(
|
136 |
-
boxes_filt,
|
|
|
137 |
|
138 |
masks, _, _ = sam_predictor.predict_torch(
|
139 |
point_coords=None,
|
@@ -142,80 +232,34 @@ def get_mask(image, label):
|
|
142 |
multimask_output=False,
|
143 |
)
|
144 |
result_mask = masks[0][0].cpu().numpy()
|
|
|
145 |
|
146 |
-
result_mask = Image.fromarray(result_mask)
|
147 |
-
|
148 |
-
return result_mask
|
149 |
|
150 |
def create_highlighted_mask(image_np, mask_np, alpha=0.5, gray_value=128):
|
151 |
-
|
152 |
-
|
153 |
if mask_np.max() <= 1.0:
|
154 |
mask_np = (mask_np * 255).astype(np.uint8)
|
155 |
mask_bool = mask_np > 128
|
156 |
-
|
157 |
image_float = image_np.astype(np.float32)
|
158 |
-
|
159 |
-
# 灰色图层
|
160 |
gray_overlay = np.full_like(image_float, gray_value, dtype=np.float32)
|
161 |
-
|
162 |
-
# 混合
|
163 |
result = image_float.copy()
|
164 |
-
result[mask_bool] = (
|
165 |
-
(1 - alpha) * image_float[mask_bool] + alpha * gray_overlay[mask_bool]
|
166 |
-
)
|
167 |
-
|
168 |
return result.astype(np.uint8)
|
169 |
|
170 |
-
hf_token = os.getenv("HF_TOKEN")
|
171 |
-
|
172 |
-
snapshot_download(repo_id="black-forest-labs/FLUX.1-Fill-dev", local_dir="./FLUX.1-Fill-dev", token=hf_token)
|
173 |
-
snapshot_download(repo_id="black-forest-labs/FLUX.1-Redux-dev", local_dir="./FLUX.1-Redux-dev", token=hf_token)
|
174 |
-
snapshot_download(repo_id="WensongSong/Insert-Anything", local_dir="./insertanything_model", token=hf_token)
|
175 |
|
|
|
|
|
|
|
|
|
|
|
176 |
|
177 |
-
|
178 |
-
|
179 |
-
|
180 |
-
|
181 |
-
"./FLUX.1-Fill-dev",
|
182 |
-
torch_dtype=dtype
|
183 |
-
).to("cuda")
|
184 |
-
|
185 |
-
pipe.load_lora_weights(
|
186 |
-
"./insertanything_model/20250321_steps5000_pytorch_lora_weights.safetensors"
|
187 |
-
)
|
188 |
-
|
189 |
-
|
190 |
-
redux = FluxPriorReduxPipeline.from_pretrained("./FLUX.1-Redux-dev").to(dtype=dtype).to("cuda")
|
191 |
-
|
192 |
-
|
193 |
-
|
194 |
-
### example #####
|
195 |
-
ref_dir='./examples/ref_image'
|
196 |
-
ref_mask_dir='./examples/ref_mask'
|
197 |
-
image_dir='./examples/source_image'
|
198 |
-
image_mask_dir='./examples/source_mask'
|
199 |
-
|
200 |
-
ref_list=[os.path.join(ref_dir,file) for file in os.listdir(ref_dir) if '.jpg' in file or '.png' in file or '.jpeg' in file ]
|
201 |
-
ref_list.sort()
|
202 |
-
|
203 |
-
ref_mask_list=[os.path.join(ref_mask_dir,file) for file in os.listdir(ref_mask_dir) if '.jpg' in file or '.png' in file or '.jpeg' in file]
|
204 |
-
ref_mask_list.sort()
|
205 |
-
|
206 |
-
image_list=[os.path.join(image_dir,file) for file in os.listdir(image_dir) if '.jpg' in file or '.png' in file or '.jpeg' in file ]
|
207 |
-
image_list.sort()
|
208 |
-
|
209 |
-
image_mask_list=[os.path.join(image_mask_dir,file) for file in os.listdir(image_mask_dir) if '.jpg' in file or '.png' in file or '.jpeg' in file]
|
210 |
-
image_mask_list.sort()
|
211 |
-
### example #####
|
212 |
-
|
213 |
|
214 |
|
215 |
@spaces.GPU
|
216 |
def run_local(base_image, base_mask, reference_image, ref_mask, seed, base_mask_option, ref_mask_option, text_prompt):
|
217 |
-
|
218 |
-
|
219 |
if base_mask_option == "Draw Mask":
|
220 |
tar_image = base_image["background"]
|
221 |
tar_mask = base_image["layers"][0]
|
@@ -250,42 +294,37 @@ def run_local(base_image, base_mask, reference_image, ref_mask, seed, base_mask_
|
|
250 |
|
251 |
if tar_mask.sum() == 0:
|
252 |
raise gr.Error('No mask for the background image.Please check mask button!')
|
253 |
-
|
254 |
if ref_mask.sum() == 0:
|
255 |
raise gr.Error('No mask for the reference image.Please check mask button!')
|
256 |
|
257 |
ref_box_yyxx = get_bbox_from_mask(ref_mask)
|
258 |
-
ref_mask_3 = np.stack([ref_mask,ref_mask,ref_mask]
|
259 |
-
masked_ref_image = ref_image * ref_mask_3 + np.ones_like(ref_image) * 255 * (1-ref_mask_3)
|
260 |
-
y1,y2,x1,x2 = ref_box_yyxx
|
261 |
-
masked_ref_image = masked_ref_image[y1:y2,x1:x2
|
262 |
-
ref_mask = ref_mask[y1:y2,x1:x2]
|
263 |
ratio = 1.3
|
264 |
masked_ref_image, ref_mask = expand_image_mask(masked_ref_image, ref_mask, ratio=ratio)
|
265 |
|
266 |
-
|
267 |
-
masked_ref_image = pad_to_square(masked_ref_image, pad_value = 255, random = False)
|
268 |
|
269 |
kernel = np.ones((7, 7), np.uint8)
|
270 |
iterations = 2
|
271 |
tar_mask = cv2.dilate(tar_mask, kernel, iterations=iterations)
|
272 |
|
273 |
-
#
|
274 |
tar_box_yyxx = get_bbox_from_mask(tar_mask)
|
275 |
tar_box_yyxx = expand_bbox(tar_mask, tar_box_yyxx, ratio=1.2)
|
276 |
|
277 |
-
tar_box_yyxx_crop =
|
278 |
-
tar_box_yyxx_crop = box2squre(tar_image, tar_box_yyxx_crop)
|
279 |
-
y1,y2,x1,x2 = tar_box_yyxx_crop
|
280 |
-
|
281 |
|
282 |
old_tar_image = tar_image.copy()
|
283 |
-
tar_image = tar_image[y1:y2,x1:x2
|
284 |
-
tar_mask = tar_mask[y1:y2,x1:x2]
|
285 |
|
286 |
H1, W1 = tar_image.shape[0], tar_image.shape[1]
|
287 |
-
# zome in
|
288 |
-
|
289 |
|
290 |
tar_mask = pad_to_square(tar_mask, pad_value=0)
|
291 |
tar_mask = cv2.resize(tar_mask, size)
|
@@ -293,19 +332,15 @@ def run_local(base_image, base_mask, reference_image, ref_mask, seed, base_mask_
|
|
293 |
masked_ref_image = cv2.resize(masked_ref_image.astype(np.uint8), size).astype(np.uint8)
|
294 |
pipe_prior_output = redux(Image.fromarray(masked_ref_image))
|
295 |
|
296 |
-
|
297 |
tar_image = pad_to_square(tar_image, pad_value=255)
|
298 |
-
|
299 |
H2, W2 = tar_image.shape[0], tar_image.shape[1]
|
300 |
-
|
301 |
tar_image = cv2.resize(tar_image, size)
|
302 |
diptych_ref_tar = np.concatenate([masked_ref_image, tar_image], axis=1)
|
303 |
|
304 |
-
|
305 |
-
tar_mask = np.stack([tar_mask,tar_mask,tar_mask],-1)
|
306 |
mask_black = np.ones_like(tar_image) * 0
|
307 |
mask_diptych = np.concatenate([mask_black, tar_mask], axis=1)
|
308 |
-
|
309 |
show_diptych_ref_tar = create_highlighted_mask(diptych_ref_tar, mask_diptych)
|
310 |
show_diptych_ref_tar = Image.fromarray(show_diptych_ref_tar)
|
311 |
|
@@ -313,8 +348,6 @@ def run_local(base_image, base_mask, reference_image, ref_mask, seed, base_mask_
|
|
313 |
mask_diptych[mask_diptych == 1] = 255
|
314 |
mask_diptych = Image.fromarray(mask_diptych)
|
315 |
|
316 |
-
|
317 |
-
|
318 |
generator = torch.Generator("cuda").manual_seed(seed)
|
319 |
edited_image = pipe(
|
320 |
image=diptych_ref_tar,
|
@@ -323,27 +356,22 @@ def run_local(base_image, base_mask, reference_image, ref_mask, seed, base_mask_
|
|
323 |
width=mask_diptych.size[0],
|
324 |
max_sequence_length=512,
|
325 |
generator=generator,
|
326 |
-
**pipe_prior_output,
|
327 |
).images[0]
|
328 |
|
329 |
-
|
330 |
-
|
331 |
width, height = edited_image.size
|
332 |
left = width // 2
|
333 |
-
|
334 |
-
top = 0
|
335 |
-
bottom = height
|
336 |
-
edited_image = edited_image.crop((left, top, right, bottom))
|
337 |
-
|
338 |
|
339 |
edited_image = np.array(edited_image)
|
340 |
-
edited_image = crop_back(edited_image, old_tar_image, np.array([H1, W1, H2, W2]), np.array(tar_box_yyxx_crop))
|
341 |
edited_image = Image.fromarray(edited_image)
|
342 |
|
343 |
if ref_mask_option != "Label to Mask":
|
344 |
return [show_diptych_ref_tar, edited_image]
|
345 |
else:
|
346 |
-
return [return_ref_mask, show_diptych_ref_tar, edited_image]
|
|
|
347 |
|
348 |
def update_ui(option):
|
349 |
if option == "Draw Mask":
|
@@ -353,8 +381,6 @@ def update_ui(option):
|
|
353 |
|
354 |
|
355 |
with gr.Blocks() as demo:
|
356 |
-
|
357 |
-
|
358 |
gr.Markdown("# Insert-Anything")
|
359 |
gr.Markdown("### Make sure to select the correct mask button!!")
|
360 |
gr.Markdown("### Click the output image to toggle between Diptych and final results!!")
|
@@ -362,42 +388,42 @@ with gr.Blocks() as demo:
|
|
362 |
with gr.Row():
|
363 |
with gr.Column(scale=1):
|
364 |
with gr.Row():
|
365 |
-
base_image = gr.ImageEditor(label="Background Image", sources="upload", type="pil",
|
366 |
-
|
367 |
-
|
368 |
-
|
369 |
-
|
370 |
-
|
371 |
with gr.Row():
|
372 |
-
base_mask_option = gr.Radio(["Draw Mask", "Upload with Mask"], label="Background Mask Input Option",
|
|
|
373 |
|
374 |
with gr.Row():
|
375 |
-
ref_image = gr.ImageEditor(label="Reference Image", sources="upload", type="pil",
|
376 |
-
|
377 |
-
|
378 |
-
|
379 |
-
|
380 |
|
381 |
with gr.Row():
|
382 |
-
ref_mask_option = gr.Radio(["Draw Mask", "Upload with Mask", "Label to Mask"],
|
383 |
-
|
384 |
with gr.Row():
|
385 |
-
text_prompt = gr.Textbox(label="Label",
|
|
|
386 |
|
387 |
with gr.Column(scale=1):
|
388 |
baseline_gallery = gr.Gallery(label='Output', show_label=True, elem_id="gallery", height=695, columns=1)
|
389 |
with gr.Accordion("Advanced Option", open=True):
|
390 |
-
seed = gr.Slider(label="Seed", minimum=-1, maximum=
|
391 |
gr.Markdown("### Guidelines")
|
392 |
gr.Markdown(" Users can try using different seeds. For example, seeds like 42 and 123456 may produce different effects.")
|
393 |
gr.Markdown(" Draw Mask means manually drawing a mask on the original image.")
|
394 |
gr.Markdown(" Upload with Mask means uploading a mask file.")
|
395 |
gr.Markdown(" Label to Mask means simply inputting a label to automatically extract the mask and obtain the result.")
|
396 |
-
|
397 |
|
398 |
run_local_button = gr.Button(value="Run")
|
399 |
|
400 |
-
#
|
401 |
num_examples = len(image_list)
|
402 |
for i in range(num_examples):
|
403 |
with gr.Row():
|
@@ -413,10 +439,10 @@ with gr.Blocks() as demo:
|
|
413 |
gr.Examples([ref_mask_list[i]], inputs=[ref_mask], examples_per_page=1, label="")
|
414 |
if i < num_examples - 1:
|
415 |
gr.HTML("<hr>")
|
416 |
-
# #### example #####
|
417 |
|
418 |
-
run_local_button.click(
|
419 |
-
|
420 |
-
|
421 |
-
|
422 |
-
|
|
|
|
1 |
+
# app.py — storage-safe + HF Hub friendly
|
2 |
+
|
3 |
import os
|
4 |
import sys
|
5 |
import cv2
|
|
|
8 |
import gradio as gr
|
9 |
from PIL import Image, ImageFilter, ImageDraw
|
10 |
|
11 |
+
# ---------- ENV & THREADS ----------
|
12 |
+
# Map a Spaces variable (no underscores allowed) to the real OpenMP var.
|
13 |
+
omp_val = os.getenv("OMP-NUM-THREADS") or os.getenv("OMPNUMTHREADS") or "2"
|
14 |
+
os.environ["OMP_NUM_THREADS"] = omp_val
|
15 |
+
try:
|
16 |
+
torch.set_num_threads(int(omp_val))
|
17 |
+
torch.set_num_interop_threads(1)
|
18 |
+
except Exception:
|
19 |
+
pass
|
20 |
+
|
21 |
+
# Send all caches to persistent storage
|
22 |
+
os.environ.setdefault("HF_HOME", "/data/.huggingface")
|
23 |
+
os.environ.setdefault("HF_HUB_CACHE", "/data/.huggingface/hub")
|
24 |
+
os.environ.setdefault("TRANSFORMERS_CACHE", "/data/.huggingface/transformers")
|
25 |
+
os.environ.setdefault("HF_DATASETS_CACHE", "/data/.huggingface/datasets")
|
26 |
+
|
27 |
+
# Disable Xet path, enable fast transfer
|
28 |
+
os.environ.setdefault("HF_HUB_DISABLE_XET", "1")
|
29 |
+
os.environ.setdefault("HF_HUB_ENABLE_HF_TRANSFER", "1")
|
30 |
+
|
31 |
+
# ---------- HUB IMPORTS ----------
|
32 |
+
from huggingface_hub import snapshot_download, hf_hub_download # noqa: E402
|
33 |
+
from diffusers import FluxFillPipeline, FluxPriorReduxPipeline # noqa: E402
|
34 |
+
|
35 |
+
import math # noqa: E402
|
36 |
+
from utils.utils import ( # noqa: E402
|
37 |
+
get_bbox_from_mask, expand_bbox, pad_to_square, box2squre, crop_back, expand_image_mask
|
38 |
+
)
|
39 |
|
40 |
+
# Optional editable installs ONLY if import fails (use requirements.txt ideally)
|
41 |
+
def _ensure_local_editable(pkg_name, rel_path):
|
42 |
+
try:
|
43 |
+
__import__(pkg_name)
|
44 |
+
except ImportError:
|
45 |
+
os.system(f"python -m pip install -e {rel_path}")
|
46 |
|
47 |
+
_ensure_local_editable("segment_anything", "segment_anything")
|
48 |
+
_ensure_local_editable("GroundingDINO", "GroundingDINO")
|
|
|
|
|
49 |
|
|
|
|
|
|
|
50 |
sys.path.append(os.path.join(os.getcwd(), "GroundingDINO"))
|
51 |
sys.path.append(os.path.join(os.getcwd(), "segment_anything"))
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
52 |
|
53 |
+
import torchvision # noqa: E402
|
54 |
+
from GroundingDINO.groundingdino.util.inference import load_model # noqa: E402
|
55 |
+
from segment_anything import build_sam, SamPredictor # noqa: E402
|
56 |
+
import spaces # noqa: E402
|
57 |
+
import GroundingDINO.groundingdino.datasets.transforms as T # noqa: E402
|
58 |
+
from GroundingDINO.groundingdino.util.utils import clean_state_dict, get_phrases_from_posmap # noqa: E402
|
59 |
|
60 |
+
# ---------- PATHS ----------
|
61 |
+
PERSIST_ROOT = "/data"
|
62 |
+
MODELS_DIR = os.path.join(PERSIST_ROOT, "models")
|
63 |
+
CKPT_DIR = os.path.join(PERSIST_ROOT, "checkpoints")
|
64 |
+
os.makedirs(MODELS_DIR, exist_ok=True)
|
65 |
+
os.makedirs(CKPT_DIR, exist_ok=True)
|
66 |
|
67 |
# GroundingDINO config and checkpoint
|
68 |
GROUNDING_DINO_CONFIG_PATH = "./GroundingDINO_SwinB.cfg.py"
|
69 |
+
GROUNDING_DINO_CHECKPOINT_PATH = os.path.join(CKPT_DIR, "groundingdino_swinb_cogcoor.pth")
|
70 |
|
71 |
# Segment-Anything checkpoint
|
72 |
SAM_ENCODER_VERSION = "vit_h"
|
73 |
+
SAM_CHECKPOINT_PATH = os.path.join(CKPT_DIR, "sam_vit_h_4b8939.pth")
|
74 |
+
|
75 |
+
# ---------- AUTH TOKEN ----------
|
76 |
+
hf_token = os.getenv("HF_TOKEN") or os.getenv("HUGGINGFACE_HUB_TOKEN")
|
77 |
+
|
78 |
+
# ---------- DOWNLOAD CHECKPOINTS (single files) ----------
|
79 |
+
# GroundingDINO ckpt (single file)
|
80 |
+
if not os.path.exists(GROUNDING_DINO_CHECKPOINT_PATH):
|
81 |
+
G_DINO_FILE = hf_hub_download(
|
82 |
+
repo_id="ShilongLiu/GroundingDINO",
|
83 |
+
filename="groundingdino_swinb_cogcoor.pth",
|
84 |
+
local_dir=CKPT_DIR,
|
85 |
+
token=hf_token,
|
86 |
+
)
|
87 |
+
if G_DINO_FILE != GROUNDING_DINO_CHECKPOINT_PATH:
|
88 |
+
# Ensure the expected path exists for later code
|
89 |
+
os.replace(G_DINO_FILE, GROUNDING_DINO_CHECKPOINT_PATH)
|
90 |
+
|
91 |
+
# SAM ckpt (single file)
|
92 |
+
if not os.path.exists(SAM_CHECKPOINT_PATH):
|
93 |
+
SAM_FILE = hf_hub_download(
|
94 |
+
repo_id="spaces/mrtlive/segment-anything-model",
|
95 |
+
filename="sam_vit_h_4b8939.pth",
|
96 |
+
local_dir=CKPT_DIR,
|
97 |
+
token=hf_token,
|
98 |
+
)
|
99 |
+
if SAM_FILE != SAM_CHECKPOINT_PATH:
|
100 |
+
os.replace(SAM_FILE, SAM_CHECKPOINT_PATH)
|
101 |
+
|
102 |
+
# ---------- DOWNLOAD MODELS (filtered snapshots into /data) ----------
|
103 |
+
FILL_DIR = os.path.join(MODELS_DIR, "FLUX.1-Fill-dev")
|
104 |
+
REDUX_DIR = os.path.join(MODELS_DIR, "FLUX.1-Redux-dev")
|
105 |
+
LORA_DIR = os.path.join(MODELS_DIR, "insertanything_model")
|
106 |
+
for path in (FILL_DIR, REDUX_DIR, LORA_DIR):
|
107 |
+
os.makedirs(path, exist_ok=True)
|
108 |
+
|
109 |
+
# Only pull what we need (weights/configs). Keep symlinks to avoid copies.
|
110 |
+
if not os.listdir(FILL_DIR):
|
111 |
+
snapshot_download(
|
112 |
+
repo_id="black-forest-labs/FLUX.1-Fill-dev",
|
113 |
+
local_dir=FILL_DIR,
|
114 |
+
local_dir_use_symlinks=True,
|
115 |
+
allow_patterns=["*.safetensors", "*.json", "*.yaml", "*.txt", "*.py", "*.model"],
|
116 |
+
token=hf_token,
|
117 |
+
)
|
118 |
+
|
119 |
+
if not os.listdir(REDUX_DIR):
|
120 |
+
snapshot_download(
|
121 |
+
repo_id="black-forest-labs/FLUX.1-Redux-dev",
|
122 |
+
local_dir=REDUX_DIR,
|
123 |
+
local_dir_use_symlinks=True,
|
124 |
+
allow_patterns=["*.safetensors", "*.json", "*.yaml", "*.txt", "*.py", "*.model"],
|
125 |
+
token=hf_token,
|
126 |
+
)
|
127 |
+
|
128 |
+
if not os.listdir(LORA_DIR):
|
129 |
+
snapshot_download(
|
130 |
+
repo_id="WensongSong/Insert-Anything",
|
131 |
+
local_dir=LORA_DIR,
|
132 |
+
local_dir_use_symlinks=True,
|
133 |
+
allow_patterns=["*.safetensors", "*.json", "*.yaml", "*.txt"],
|
134 |
+
token=hf_token,
|
135 |
+
)
|
136 |
+
|
137 |
+
# ---------- BUILD MODELS ----------
|
138 |
+
# GroundingDINO
|
139 |
+
groundingdino_model = load_model(
|
140 |
+
model_config_path=GROUNDING_DINO_CONFIG_PATH,
|
141 |
+
model_checkpoint_path=GROUNDING_DINO_CHECKPOINT_PATH,
|
142 |
+
device="cuda"
|
143 |
+
)
|
144 |
|
145 |
+
# SAM + Predictor
|
|
|
|
|
146 |
sam = build_sam(checkpoint=SAM_CHECKPOINT_PATH)
|
147 |
sam.to(device="cuda")
|
148 |
sam_predictor = SamPredictor(sam)
|
149 |
|
150 |
+
# Diffusers
|
151 |
+
dtype = torch.bfloat16
|
152 |
+
size = (768, 768)
|
153 |
+
|
154 |
+
pipe = FluxFillPipeline.from_pretrained(
|
155 |
+
FILL_DIR,
|
156 |
+
torch_dtype=dtype
|
157 |
+
).to("cuda")
|
158 |
+
|
159 |
+
pipe.load_lora_weights(
|
160 |
+
os.path.join(LORA_DIR, "20250321_steps5000_pytorch_lora_weights.safetensors")
|
161 |
+
)
|
162 |
|
163 |
+
redux = FluxPriorReduxPipeline.from_pretrained(REDUX_DIR).to(dtype=dtype).to("cuda")
|
164 |
+
|
165 |
+
# ---------- APP LOGIC ----------
|
166 |
+
def transform_image(image_pil):
|
167 |
transform = T.Compose(
|
168 |
[
|
169 |
T.RandomResize([800], max_size=1333),
|
|
|
176 |
|
177 |
|
178 |
def get_grounding_output(model, image, caption, box_threshold=0.25, text_threshold=0.25, with_logits=True):
|
179 |
+
caption = caption.lower().strip()
|
|
|
180 |
if not caption.endswith("."):
|
181 |
caption = caption + "."
|
|
|
182 |
with torch.no_grad():
|
183 |
outputs = model(image[None], captions=[caption])
|
184 |
logits = outputs["pred_logits"].cpu().sigmoid()[0] # (nq, 256)
|
185 |
+
boxes = outputs["pred_boxes"].cpu()[0] # (nq, 4)
|
|
|
186 |
|
187 |
# filter output
|
188 |
+
filt_mask = logits.max(dim=1)[0] > box_threshold
|
189 |
+
logits_filt = logits[filt_mask]
|
190 |
+
boxes_filt = boxes[filt_mask]
|
|
|
|
|
|
|
191 |
|
192 |
# get phrase
|
193 |
tokenlizer = model.tokenizer
|
194 |
tokenized = tokenlizer(caption)
|
195 |
+
pred_phrases, scores = [], []
|
|
|
|
|
196 |
for logit, box in zip(logits_filt, boxes_filt):
|
197 |
+
pred_phrase = get_phrases_from_posmap(logit > text_threshold, tokenized, tokenlizer)
|
198 |
+
pred_phrases.append(pred_phrase + f"({str(logit.max().item())[:4]})" if with_logits else pred_phrase)
|
|
|
|
|
|
|
|
|
|
|
199 |
scores.append(logit.max().item())
|
|
|
200 |
return boxes_filt, torch.Tensor(scores), pred_phrases
|
201 |
|
202 |
|
203 |
def get_mask(image, label):
|
204 |
global groundingdino_model, sam_predictor
|
|
|
|
|
205 |
image_pil = image.convert("RGB")
|
206 |
transformed_image = transform_image(image_pil)
|
207 |
|
|
|
208 |
boxes_filt, scores, pred_phrases = get_grounding_output(
|
209 |
groundingdino_model, transformed_image, label
|
210 |
)
|
211 |
|
212 |
+
W, H = image_pil.size
|
|
|
|
|
|
|
213 |
for i in range(boxes_filt.size(0)):
|
214 |
boxes_filt[i] = boxes_filt[i] * torch.Tensor([W, H, W, H])
|
215 |
boxes_filt[i][:2] -= boxes_filt[i][2:] / 2
|
216 |
boxes_filt[i][2:] += boxes_filt[i][:2]
|
|
|
217 |
boxes_filt = boxes_filt.cpu()
|
218 |
|
219 |
+
nms_idx = torchvision.ops.nms(boxes_filt, scores, 0.8).numpy().tolist()
|
|
|
|
|
|
|
220 |
boxes_filt = boxes_filt[nms_idx]
|
|
|
|
|
|
|
|
|
|
|
221 |
|
222 |
+
image_np = np.array(image_pil)
|
223 |
+
sam_predictor.set_image(image_np)
|
224 |
transformed_boxes = sam_predictor.transform.apply_boxes_torch(
|
225 |
+
boxes_filt, image_np.shape[:2]
|
226 |
+
).to("cuda")
|
227 |
|
228 |
masks, _, _ = sam_predictor.predict_torch(
|
229 |
point_coords=None,
|
|
|
232 |
multimask_output=False,
|
233 |
)
|
234 |
result_mask = masks[0][0].cpu().numpy()
|
235 |
+
return Image.fromarray(result_mask)
|
236 |
|
|
|
|
|
|
|
237 |
|
238 |
def create_highlighted_mask(image_np, mask_np, alpha=0.5, gray_value=128):
|
|
|
|
|
239 |
if mask_np.max() <= 1.0:
|
240 |
mask_np = (mask_np * 255).astype(np.uint8)
|
241 |
mask_bool = mask_np > 128
|
|
|
242 |
image_float = image_np.astype(np.float32)
|
|
|
|
|
243 |
gray_overlay = np.full_like(image_float, gray_value, dtype=np.float32)
|
|
|
|
|
244 |
result = image_float.copy()
|
245 |
+
result[mask_bool] = (1 - alpha) * image_float[mask_bool] + alpha * gray_overlay[mask_bool]
|
|
|
|
|
|
|
246 |
return result.astype(np.uint8)
|
247 |
|
|
|
|
|
|
|
|
|
|
|
248 |
|
249 |
+
# ---------- EXAMPLES ----------
|
250 |
+
ref_dir = './examples/ref_image'
|
251 |
+
ref_mask_dir = './examples/ref_mask'
|
252 |
+
image_dir = './examples/source_image'
|
253 |
+
image_mask_dir = './examples/source_mask'
|
254 |
|
255 |
+
ref_list = sorted([os.path.join(ref_dir, f) for f in os.listdir(ref_dir) if f.lower().endswith((".jpg", ".png", ".jpeg"))])
|
256 |
+
ref_mask_list = sorted([os.path.join(ref_mask_dir, f) for f in os.listdir(ref_mask_dir) if f.lower().endswith((".jpg", ".png", ".jpeg"))])
|
257 |
+
image_list = sorted([os.path.join(image_dir, f) for f in os.listdir(image_dir) if f.lower().endswith((".jpg", ".png", ".jpeg"))])
|
258 |
+
image_mask_list = sorted([os.path.join(image_mask_dir, f) for f in os.listdir(image_mask_dir) if f.lower().endswith((".jpg", ".png", ".jpeg"))])
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
259 |
|
260 |
|
261 |
@spaces.GPU
|
262 |
def run_local(base_image, base_mask, reference_image, ref_mask, seed, base_mask_option, ref_mask_option, text_prompt):
|
|
|
|
|
263 |
if base_mask_option == "Draw Mask":
|
264 |
tar_image = base_image["background"]
|
265 |
tar_mask = base_image["layers"][0]
|
|
|
294 |
|
295 |
if tar_mask.sum() == 0:
|
296 |
raise gr.Error('No mask for the background image.Please check mask button!')
|
|
|
297 |
if ref_mask.sum() == 0:
|
298 |
raise gr.Error('No mask for the reference image.Please check mask button!')
|
299 |
|
300 |
ref_box_yyxx = get_bbox_from_mask(ref_mask)
|
301 |
+
ref_mask_3 = np.stack([ref_mask, ref_mask, ref_mask], -1)
|
302 |
+
masked_ref_image = ref_image * ref_mask_3 + np.ones_like(ref_image) * 255 * (1 - ref_mask_3)
|
303 |
+
y1, y2, x1, x2 = ref_box_yyxx
|
304 |
+
masked_ref_image = masked_ref_image[y1:y2, x1:x2, :]
|
305 |
+
ref_mask = ref_mask[y1:y2, x1:x2]
|
306 |
ratio = 1.3
|
307 |
masked_ref_image, ref_mask = expand_image_mask(masked_ref_image, ref_mask, ratio=ratio)
|
308 |
|
309 |
+
masked_ref_image = pad_to_square(masked_ref_image, pad_value=255, random=False)
|
|
|
310 |
|
311 |
kernel = np.ones((7, 7), np.uint8)
|
312 |
iterations = 2
|
313 |
tar_mask = cv2.dilate(tar_mask, kernel, iterations=iterations)
|
314 |
|
315 |
+
# zoom in
|
316 |
tar_box_yyxx = get_bbox_from_mask(tar_mask)
|
317 |
tar_box_yyxx = expand_bbox(tar_mask, tar_box_yyxx, ratio=1.2)
|
318 |
|
319 |
+
tar_box_yyxx_crop = expand_bbox(tar_image, tar_box_yyxx, ratio=2)
|
320 |
+
tar_box_yyxx_crop = box2squre(tar_image, tar_box_yyxx_crop) # crop box
|
321 |
+
y1, y2, x1, x2 = tar_box_yyxx_crop
|
|
|
322 |
|
323 |
old_tar_image = tar_image.copy()
|
324 |
+
tar_image = tar_image[y1:y2, x1:x2, :]
|
325 |
+
tar_mask = tar_mask[y1:y2, x1:x2]
|
326 |
|
327 |
H1, W1 = tar_image.shape[0], tar_image.shape[1]
|
|
|
|
|
328 |
|
329 |
tar_mask = pad_to_square(tar_mask, pad_value=0)
|
330 |
tar_mask = cv2.resize(tar_mask, size)
|
|
|
332 |
masked_ref_image = cv2.resize(masked_ref_image.astype(np.uint8), size).astype(np.uint8)
|
333 |
pipe_prior_output = redux(Image.fromarray(masked_ref_image))
|
334 |
|
|
|
335 |
tar_image = pad_to_square(tar_image, pad_value=255)
|
|
|
336 |
H2, W2 = tar_image.shape[0], tar_image.shape[1]
|
|
|
337 |
tar_image = cv2.resize(tar_image, size)
|
338 |
diptych_ref_tar = np.concatenate([masked_ref_image, tar_image], axis=1)
|
339 |
|
340 |
+
tar_mask = np.stack([tar_mask, tar_mask, tar_mask], -1)
|
|
|
341 |
mask_black = np.ones_like(tar_image) * 0
|
342 |
mask_diptych = np.concatenate([mask_black, tar_mask], axis=1)
|
343 |
+
|
344 |
show_diptych_ref_tar = create_highlighted_mask(diptych_ref_tar, mask_diptych)
|
345 |
show_diptych_ref_tar = Image.fromarray(show_diptych_ref_tar)
|
346 |
|
|
|
348 |
mask_diptych[mask_diptych == 1] = 255
|
349 |
mask_diptych = Image.fromarray(mask_diptych)
|
350 |
|
|
|
|
|
351 |
generator = torch.Generator("cuda").manual_seed(seed)
|
352 |
edited_image = pipe(
|
353 |
image=diptych_ref_tar,
|
|
|
356 |
width=mask_diptych.size[0],
|
357 |
max_sequence_length=512,
|
358 |
generator=generator,
|
359 |
+
**pipe_prior_output,
|
360 |
).images[0]
|
361 |
|
|
|
|
|
362 |
width, height = edited_image.size
|
363 |
left = width // 2
|
364 |
+
edited_image = edited_image.crop((left, 0, width, height))
|
|
|
|
|
|
|
|
|
365 |
|
366 |
edited_image = np.array(edited_image)
|
367 |
+
edited_image = crop_back(edited_image, old_tar_image, np.array([H1, W1, H2, W2]), np.array(tar_box_yyxx_crop))
|
368 |
edited_image = Image.fromarray(edited_image)
|
369 |
|
370 |
if ref_mask_option != "Label to Mask":
|
371 |
return [show_diptych_ref_tar, edited_image]
|
372 |
else:
|
373 |
+
return [return_ref_mask, show_diptych_ref_tar, edited_image]
|
374 |
+
|
375 |
|
376 |
def update_ui(option):
|
377 |
if option == "Draw Mask":
|
|
|
381 |
|
382 |
|
383 |
with gr.Blocks() as demo:
|
|
|
|
|
384 |
gr.Markdown("# Insert-Anything")
|
385 |
gr.Markdown("### Make sure to select the correct mask button!!")
|
386 |
gr.Markdown("### Click the output image to toggle between Diptych and final results!!")
|
|
|
388 |
with gr.Row():
|
389 |
with gr.Column(scale=1):
|
390 |
with gr.Row():
|
391 |
+
base_image = gr.ImageEditor(label="Background Image", sources="upload", type="pil",
|
392 |
+
brush=gr.Brush(colors=["#FFFFFF"], default_size=30, color_mode="fixed"),
|
393 |
+
layers=False, interactive=True)
|
394 |
+
base_mask = gr.ImageEditor(label="Background Mask", sources="upload", type="pil",
|
395 |
+
layers=False, brush=False, eraser=False)
|
|
|
396 |
with gr.Row():
|
397 |
+
base_mask_option = gr.Radio(["Draw Mask", "Upload with Mask"], label="Background Mask Input Option",
|
398 |
+
value="Upload with Mask")
|
399 |
|
400 |
with gr.Row():
|
401 |
+
ref_image = gr.ImageEditor(label="Reference Image", sources="upload", type="pil",
|
402 |
+
brush=gr.Brush(colors=["#FFFFFF"], default_size=30, color_mode="fixed"),
|
403 |
+
layers=False, interactive=True)
|
404 |
+
ref_mask = gr.ImageEditor(label="Reference Mask", sources="upload", type="pil",
|
405 |
+
layers=False, brush=False, eraser=False)
|
406 |
|
407 |
with gr.Row():
|
408 |
+
ref_mask_option = gr.Radio(["Draw Mask", "Upload with Mask", "Label to Mask"],
|
409 |
+
label="Reference Mask Input Option", value="Upload with Mask")
|
410 |
with gr.Row():
|
411 |
+
text_prompt = gr.Textbox(label="Label",
|
412 |
+
placeholder="Enter the category of the reference object, e.g., car, dress, toy, etc.")
|
413 |
|
414 |
with gr.Column(scale=1):
|
415 |
baseline_gallery = gr.Gallery(label='Output', show_label=True, elem_id="gallery", height=695, columns=1)
|
416 |
with gr.Accordion("Advanced Option", open=True):
|
417 |
+
seed = gr.Slider(label="Seed", minimum=-1, maximum=999_999_999, step=1, value=666)
|
418 |
gr.Markdown("### Guidelines")
|
419 |
gr.Markdown(" Users can try using different seeds. For example, seeds like 42 and 123456 may produce different effects.")
|
420 |
gr.Markdown(" Draw Mask means manually drawing a mask on the original image.")
|
421 |
gr.Markdown(" Upload with Mask means uploading a mask file.")
|
422 |
gr.Markdown(" Label to Mask means simply inputting a label to automatically extract the mask and obtain the result.")
|
|
|
423 |
|
424 |
run_local_button = gr.Button(value="Run")
|
425 |
|
426 |
+
# examples
|
427 |
num_examples = len(image_list)
|
428 |
for i in range(num_examples):
|
429 |
with gr.Row():
|
|
|
439 |
gr.Examples([ref_mask_list[i]], inputs=[ref_mask], examples_per_page=1, label="")
|
440 |
if i < num_examples - 1:
|
441 |
gr.HTML("<hr>")
|
|
|
442 |
|
443 |
+
run_local_button.click(
|
444 |
+
fn=run_local,
|
445 |
+
inputs=[base_image, base_mask, ref_image, ref_mask, seed, base_mask_option, ref_mask_option, text_prompt],
|
446 |
+
outputs=[baseline_gallery]
|
447 |
+
)
|
448 |
+
demo.launch()
|