Spaces:
Runtime error
Runtime error
File size: 12,491 Bytes
6c016cc |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 |
# --------------------------------------------------------
# Semantic-SAM: Segment and Recognize Anything at Any Granularity
# Copyright (c) 2023 Microsoft
# Licensed under The MIT License [see LICENSE for details]
# Written by Hao Zhang ([email protected])
# --------------------------------------------------------
import torch
import numpy as np
from torchvision import transforms
from task_adapter.utils.visualizer import Visualizer
from typing import Tuple
from PIL import Image
from detectron2.data import MetadataCatalog
metadata = MetadataCatalog.get('coco_2017_train_panoptic')
def interactive_infer_image(model, image,all_classes,all_parts, thresh,text_size,hole_scale,island_scale,semantic, refimg=None, reftxt=None, audio_pth=None, video_pth=None, label_mode='1', alpha=0.1, anno_mode=['Mask']):
t = []
t.append(transforms.Resize(int(text_size), interpolation=Image.BICUBIC))
transform1 = transforms.Compose(t)
image_ori = transform1(image['image'])
mask_ori = transform1(image['mask'])
width = image_ori.size[0]
height = image_ori.size[1]
image_ori = np.asarray(image_ori)
images = torch.from_numpy(image_ori.copy()).permute(2,0,1).cuda()
all_classes, all_parts=all_classes.strip().strip("\"[]").split(':'),all_parts.strip().strip("\"[]").split(':')
data = {"image": images, "height": height, "width": width}
mask_ori = np.asarray(mask_ori)[:,:,0:1].copy()
mask_ori = torch.from_numpy(mask_ori).permute(2,0,1)[0]
points=mask_ori.nonzero().float().to(images.device)
if len(points)==0:
point_=point=points.new_tensor([[0.5,0.5,0.006,0.006]])
else:
point_=points.mean(0)[None]
point=point_.clone()
point[0, 0] = point_[0, 0] / mask_ori.shape[0]
point[0, 1] = point_[0, 1] / mask_ori.shape[1]
point = point[:, [1, 0]]
point=torch.cat([point,points.new_tensor([[0.005,0.005]])],dim=-1)
data['targets'] = [dict()]
data['targets'][0]['points']=point
data['targets'][0]['pb']=point.new_tensor([0.])
batch_inputs = [data]
masks,ious = model.model.evaluate_demo(batch_inputs,all_classes,all_parts)
pred_masks_poses = masks
reses=[]
ious=ious[0,0]
ids=torch.argsort(ious,descending=True)
text_res=''
try:
thresh=float(thresh)
except Exception:
thresh=0.0
mask_ls=[]
ious_res=[]
areas=[]
for i,(pred_masks_pos,iou) in enumerate(zip(pred_masks_poses[ids],ious[ids])):
iou=round(float(iou),2)
texts=f'{iou}'
mask=(pred_masks_pos>0.0).cpu().numpy()
area=mask.sum()
conti=False
if iou<thresh:
conti=True
for m in mask_ls:
if np.logical_and(mask,m).sum()/np.logical_or(mask,m).sum()>0.95:
conti=True
break
if i == len(pred_masks_poses[ids])-1 and mask_ls==[]:
conti=False
if conti:
continue
ious_res.append(iou)
mask_ls.append(mask)
areas.append(area)
mask,_=remove_small_regions(mask,int(hole_scale),mode="holes")
mask,_=remove_small_regions(mask,int(island_scale),mode="islands")
mask=(mask).astype(np.float)
out_txt = texts
visual = Visualizer(image_ori, metadata=metadata)
color=[0.,0.,1.0]
# demo = visual.draw_binary_mask(mask, color=color, text=texts)
demo = visual.draw_binary_mask_with_number(mask, text=str(label), label_mode=label_mode, alpha=alpha, anno_mode=anno_mode)
res = demo.get_image()
point_x0=max(0,int(point_[0, 1])-3)
point_x1=min(mask_ori.shape[1],int(point_[0, 1])+3)
point_y0 = max(0, int(point_[0, 0]) - 3)
point_y1 = min(mask_ori.shape[0], int(point_[0, 0]) + 3)
# res[point_y0:point_y1,point_x0:point_x1,0]=255
# res[point_y0:point_y1,point_x0:point_x1,1]=0
# res[point_y0:point_y1,point_x0:point_x1,2]=0
reses.append(Image.fromarray(res))
text_res=text_res+';'+out_txt
ids=list(torch.argsort(torch.tensor(areas),descending=False))
ids = [int(i) for i in ids]
torch.cuda.empty_cache()
return reses,[reses[i] for i in ids]
def interactive_infer_image_3l(model, image,all_classes,all_parts, thresh,text_size,hole_scale,island_scale,semantic, refimg=None, reftxt=None, audio_pth=None, video_pth=None):
t = []
t.append(transforms.Resize(int(text_size), interpolation=Image.BICUBIC))
transform1 = transforms.Compose(t)
image_ori = transform1(image['image'])
mask_ori = transform1(image['mask'])
width = image_ori.size[0]
height = image_ori.size[1]
image_ori = np.asarray(image_ori)
images = torch.from_numpy(image_ori.copy()).permute(2,0,1).cuda()
all_classes, all_parts=all_classes.strip().strip("\"[]").split(':'),all_parts.strip().strip("\"[]").split(':')
data = {"image": images, "height": height, "width": width}
mask_ori = np.asarray(mask_ori)[:,:,0:1].copy()
mask_ori = torch.from_numpy(mask_ori).permute(2,0,1)[0]
points=mask_ori.nonzero().float().to(images.device)
if len(points)==0:
point_=point=points.new_tensor([[0.5,0.5,0.006,0.006]])
else:
point_=points.mean(0)[None]
point=point_.clone()
point[0, 0] = point_[0, 0] / mask_ori.shape[0]
point[0, 1] = point_[0, 1] / mask_ori.shape[1]
point = point[:, [1, 0]]
point=torch.cat([point,points.new_tensor([[0.005,0.005]])],dim=-1)
data['targets'] = [dict()]
data['targets'][0]['points']=point
data['targets'][0]['pb']=point.new_tensor([0.])
batch_inputs = [data]
masks, ious, pred_class, pred_class_score = model.model.evaluate_demo(batch_inputs,all_classes,all_parts, level=[0,1,2])
pred_masks_poses = masks
reses=[]
ious=ious[0,0]
ids=torch.argsort(ious,descending=True)
text_res=''
try:
thresh=float(thresh)
except Exception:
thresh=0.0
mask_ls=[]
ious_res=[]
areas=[]
new_pred_class = []
new_pred_class_score = []
for i in ids:
new_pred_class_score.append(pred_class_score[i])
new_pred_class.append(pred_class[i])
# import ipdb; ipdb.set_trace()
for i,(pred_masks_pos,iou, cls_name, cls_score) in enumerate(zip(pred_masks_poses[ids],ious[ids], new_pred_class, new_pred_class_score)):
iou=round(float(iou),2)
texts=f'{iou}_{cls_name}_{cls_score}'
mask=(pred_masks_pos>0.0).cpu().numpy()
area=mask.sum()
conti=False
if iou<thresh:
conti=True
for m in mask_ls:
if np.logical_and(mask,m).sum()/np.logical_or(mask,m).sum()>0.95:
conti=True
break
if i == len(pred_masks_poses[ids])-1 and mask_ls==[]:
conti=False
if conti:
continue
ious_res.append(iou)
mask_ls.append(mask)
areas.append(area)
mask,_=remove_small_regions(mask,int(hole_scale),mode="holes")
mask,_=remove_small_regions(mask,int(island_scale),mode="islands")
mask=(mask).astype(np.float)
out_txt = texts
visual = Visualizer(image_ori, metadata=metadata)
color=[0.,0.,1.0]
demo = visual.draw_binary_mask(mask, color=color, text=texts)
res = demo.get_image()
point_x0=max(0,int(point_[0, 1])-3)
point_x1=min(mask_ori.shape[1],int(point_[0, 1])+3)
point_y0 = max(0, int(point_[0, 0]) - 3)
point_y1 = min(mask_ori.shape[0], int(point_[0, 0]) + 3)
res[point_y0:point_y1,point_x0:point_x1,0]=255
res[point_y0:point_y1,point_x0:point_x1,1]=0
res[point_y0:point_y1,point_x0:point_x1,2]=0
reses.append(Image.fromarray(res))
text_res=text_res+';'+out_txt
ids=list(torch.argsort(torch.tensor(areas),descending=False))
ids = [int(i) for i in ids]
torch.cuda.empty_cache()
return reses,[reses[i] for i in ids]
def interactive_infer_image_semantic(model, image,all_classes,all_parts, thresh,text_size,hole_scale,island_scale,semantic, refimg=None, reftxt=None, audio_pth=None, video_pth=None):
t = []
t.append(transforms.Resize(int(text_size), interpolation=Image.BICUBIC))
transform1 = transforms.Compose(t)
image_ori = transform1(image['image'])
mask_ori = transform1(image['mask'])
width = image_ori.size[0]
height = image_ori.size[1]
image_ori = np.asarray(image_ori)
images = torch.from_numpy(image_ori.copy()).permute(2,0,1).cuda()
all_classes, all_parts=all_classes.strip().strip("\"[]").split(':'),all_parts.strip().strip("\"[]").split(':')
data = {"image": images, "height": height, "width": width}
mask_ori = np.asarray(mask_ori)[:,:,0:1].copy()
mask_ori = torch.from_numpy(mask_ori).permute(2,0,1)[0]
points=mask_ori.nonzero().float().to(images.device)
if len(points)==0:
point_=point=points.new_tensor([[0.5,0.5,0.006,0.006]])
else:
point_=points.mean(0)[None]
point=point_.clone()
point[0, 0] = point_[0, 0] / mask_ori.shape[0]
point[0, 1] = point_[0, 1] / mask_ori.shape[1]
point = point[:, [1, 0]]
point=torch.cat([point,points.new_tensor([[0.005,0.005]])],dim=-1)
data['targets'] = [dict()]
data['targets'][0]['points']=point
data['targets'][0]['pb']=point.new_tensor([0.])
data['targets'][0]['pb']=point.new_tensor([1.])
batch_inputs = [data]
masks,ious = model.model.evaluate_demo(batch_inputs,all_classes,all_parts)
pred_masks_poses = masks
reses=[]
ious=ious[0,0]
ids=torch.argsort(ious,descending=True)
text_res=''
try:
thresh=float(thresh)
except Exception:
thresh=0.0
mask_ls=[]
ious_res=[]
areas=[]
for i,(pred_masks_pos,iou) in enumerate(zip(pred_masks_poses[ids],ious[ids])):
iou=round(float(iou),2)
texts=f'{iou}'
mask=(pred_masks_pos>0.0).cpu().numpy()
area=mask.sum()
conti=False
if iou<thresh:
conti=True
for m in mask_ls:
if np.logical_and(mask,m).sum()/np.logical_or(mask,m).sum()>0.95:
conti=True
break
if i == len(pred_masks_poses[ids])-1 and mask_ls==[]:
conti=False
if conti:
continue
ious_res.append(iou)
mask_ls.append(mask)
areas.append(area)
mask,_=remove_small_regions(mask,int(hole_scale),mode="holes")
mask,_=remove_small_regions(mask,int(island_scale),mode="islands")
mask=(mask).astype(np.float)
out_txt = texts
visual = Visualizer(image_ori, metadata=metadata)
color=[0.,0.,1.0]
demo = visual.draw_binary_mask(mask, color=color, text=texts)
res = demo.get_image()
point_x0=max(0,int(point_[0, 1])-3)
point_x1=min(mask_ori.shape[1],int(point_[0, 1])+3)
point_y0 = max(0, int(point_[0, 0]) - 3)
point_y1 = min(mask_ori.shape[0], int(point_[0, 0]) + 3)
res[point_y0:point_y1,point_x0:point_x1,0]=255
res[point_y0:point_y1,point_x0:point_x1,1]=0
res[point_y0:point_y1,point_x0:point_x1,2]=0
reses.append(Image.fromarray(res))
text_res=text_res+';'+out_txt
ids=list(torch.argsort(torch.tensor(areas),descending=False))
ids = [int(i) for i in ids]
torch.cuda.empty_cache()
return reses,[reses[i] for i in ids]
def remove_small_regions(
mask: np.ndarray, area_thresh: float, mode: str
) -> Tuple[np.ndarray, bool]:
"""
Removes small disconnected regions and holes in a mask. Returns the
mask and an indicator of if the mask has been modified.
"""
import cv2 # type: ignore
assert mode in ["holes", "islands"]
correct_holes = mode == "holes"
working_mask = (correct_holes ^ mask).astype(np.uint8)
n_labels, regions, stats, _ = cv2.connectedComponentsWithStats(working_mask, 8)
sizes = stats[:, -1][1:] # Row 0 is background label
small_regions = [i + 1 for i, s in enumerate(sizes) if s < area_thresh]
if len(small_regions) == 0:
return mask, False
fill_labels = [0] + small_regions
if not correct_holes:
fill_labels = [i for i in range(n_labels) if i not in fill_labels]
# If every region is below threshold, keep largest
if len(fill_labels) == 0:
fill_labels = [int(np.argmax(sizes)) + 1]
mask = np.isin(regions, fill_labels)
return mask, True |