开发 Vulkan 的时候,需要使用 glslangValidator 编译 Vulkan 代码。
如果是 ubuntu 19.10 版本,可以直接执行:
如果低于这个版本,则只能从源代码编译安装了,如下:
智障儿童欢乐多
开发 Vulkan 的时候,需要使用 glslangValidator 编译 Vulkan 代码。
如果是 ubuntu 19.10 版本,可以直接执行:
1 2 3 |
$ sudo apt-get install vulkan-tools # 目前ubuntu 20.04 最新需要使用如下命令 sudo apt-get install glslang-tools |
如果低于这个版本,则只能从源代码编译安装了,如下:
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 |
# 编译安装 glslang $ git clone https://github.com/KhronosGroup/glslang.git # 也可本站下载一份拷贝 wget https://www.mobibrw.com/wp-content/uploads/2018/12/glslang.zip $ cd glslang # 目前(2018.12.17)的正式版,最稳定的版本,试过最新的版本,编译部分代码存在问题 $ git checkout 7.10.2984 # 下载 spirv-tools 部分的功能代码 $ python update_glslang_sources.py $ mkdir build $ cd build $ cmake .. $ make $ sudo make install |
安装依赖
1 2 3 4 5 6 7 |
$ python -m pip install --upgrade pip --user $ python -m pip install Pillow --user $ python -m pip install Numpy --user $ python -m pip install OpenCV-Python --user |
源代码:
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 |
#!/usr/bin/python # -*- coding: UTF-8 -*- import os import sys if sys.version_info < (3, 0): import Tkinter as tk # 导入 Tkinter 库 from tkFileDialog import askopenfilename, askdirectory else : import tkinter as tk # 导入 Tkinter 库 from tkinter.filedialog import askopenfilename, askdirectory from PIL import Image, ImageTk, ImageDraw from time import sleep import numpy as np import cv2 as cv import collections DEF_WIDTH = 1080 DEF_HEIGHT = 720 IMAGE_HEIGHT = 720 FRAME_LEFT_WIDTH = 360 # 太小的选定区域我们需要丢弃,防止误操作 MINI_RECT_AREA = 20 class RawImageEditor: def __init__(self, win, img, rects): #变量X和Y用来记录鼠标左键按下的位置 self.X = tk.IntVar(value=0) self.Y = tk.IntVar(value=0) self.sel = False self.lastDraw = None self.lastDraws = [] self.imageScale = 1.0 self.dispWidth = DEF_WIDTH # 图片显示区域的最大高度,宽度 self.dispHeight = DEF_HEIGHT self.rawImage = img self.calcImageScale(self.rawImage) self.dispWidth = int(self.imageScale * self.rawImage.width) self.dispHeight = int(self.imageScale * self.rawImage.height) # 图片缩放 self.dispImage = self.rawImage.resize((self.dispWidth, self.dispHeight)) # 选择区域 self.selPositions = [] for r in rects : self.selPositions.append((r[0] * self.imageScale, r[1] * self.imageScale, r[2] * self.imageScale, r[3] * self.imageScale)) #创建顶级组件容器 self.top = tk.Toplevel(win, width=self.dispWidth, height=self.dispHeight) #不显示最大化、最小化按钮 self.top.overrideredirect(True) #center window on desktop nScreenWid, nScreenHei = self.top.maxsize() winPos = '%sx%s+%s+%s' % (int(self.top.winfo_reqwidth()), int(self.top.winfo_reqheight()), int((nScreenWid - self.top.winfo_reqwidth())/2), int((nScreenHei - self.top.winfo_reqheight())/2)) self.top.geometry(winPos) # Make topLevelWindow remain on top until destroyed, or attribute changes. self.top.attributes('-topmost', 'true') self.canvas = tk.Canvas(self.top, bg='white', width=self.dispWidth, height=self.dispHeight) self.tkImage = ImageTk.PhotoImage(self.dispImage) self.canvas.create_image(self.dispWidth//2, self.dispHeight//2, image=self.tkImage) for r in self.selPositions : draw = self.canvas.create_rectangle(r[0], r[1], r[2], r[3], outline='green') self.lastDraws.append(draw) #鼠标左键按下的位置 def onLeftButtonDown(event): self.X.set(event.x) self.Y.set(event.y) #开始截图 self.sel = True #重新绘制已经选择的区域 for draw in self.lastDraws : self.canvas.delete(draw) self.lastDraws = [] for r in self.selPositions : draw = self.canvas.create_rectangle(r[0], r[1], r[2], r[3], outline='green') self.lastDraws.append(draw) self.canvas.bind('<Button-1>', onLeftButtonDown) #鼠标左键移动,显示选取的区域 def onLeftButtonMove(event): if not self.sel: return try: #删除刚画完的图形,要不然鼠标移动的时候是黑乎乎的一片矩形 self.canvas.delete(self.lastDraw) except Exception as e: pass self.lastDraw = self.canvas.create_rectangle(self.X.get(), self.Y.get(), event.x, event.y, outline='green') self.canvas.bind('<B1-Motion>', onLeftButtonMove) #获取鼠标左键抬起的位置,保存区域截图 def onLeftButtonUp(event): if not self.sel: return self.sel = False sleep(0.1) #考虑鼠标左键从右下方按下而从左上方抬起的截图 left, right = sorted([self.X.get(), event.x]) top, bottom = sorted([self.Y.get(), event.y]) if (right - left) * (bottom - top) > MINI_RECT_AREA : self.selPositions.append((left,top,right,bottom)) #self.top.destroy() #鼠标右键按下 def onRightButtonDown(event): self.sel = False self.top.destroy() self.canvas.bind('<Button-2>', onRightButtonDown) self.canvas.bind('<ButtonRelease-1>', onLeftButtonUp) self.canvas.pack(fill=tk.BOTH, expand=tk.YES) def calcImageScale(self, image) : w = image.width h = image.height self.imageScale = 1.0 # 计算最小的缩放比例,保证原始宽高比 if w > self.dispWidth and h > self.dispHeight : ws = self.dispWidth * 1.0 / w hs = self.dispHeight * 1.0 / h if ws < hs : self.imageScale = ws else : self.imageScale = hs elif w > self.dispWidth and h < self.dispHeight : self.imageScale = self.dispWidth * 1.0 / w elif w < self.dispWidth and h > self.dispHeight : self.imageScale = self.dispHeight * 1.0 / h def waitForWindow(self, win) : win.wait_window(self.top) def selectedPositions(self) : # 转换为原始像素位置 realPos = [] for r in self.selPositions : realPos.append((r[0] / self.imageScale, r[1] / self.imageScale, r[2] / self.imageScale, r[3] / self.imageScale)) return realPos class MainWin(tk.Tk): def __init__(self): if sys.version_info >= (3, 0): super().__init__() else : tk.Tk.__init__(self) self.title('图像处理工具') self.geometry('{}x{}'.format(DEF_WIDTH, DEF_HEIGHT)) self.rawImagePath = '' self.rawImage = None # self.rawImage 原始图像,未经过缩放处理 self.transRawImage = None # self.transRawImage 经过转换处理之后的原始图像,没有经过缩放处理 self.dispImage = None # self.dispImage 显示图像,可能经过缩放处理 self.imageScale = 1.0 # 图片缩放比例,根据缩放比例进行显示的时候的缩放处理,后期选择区域的时候,需要进行缩放还原 self.leftFrameWidth = FRAME_LEFT_WIDTH self.frameDispHeight = DEF_HEIGHT # 整个窗口的高度 self.labelTextHeight = 18 # 文本标签的高度 self.btnHeight = 30 # 按钮的高度 self.brightnessScale = tk.StringVar() # 亮度比 self.brightnessScale.set('1.0') self.defSavePath = '' # 默认保存路径 self.imageDispWidth = IMAGE_HEIGHT # 图片显示区域的最大高度,宽度 self.imageDispHeight = self.frameDispHeight / 2 - self.labelTextHeight * 2 self.liRect = collections.OrderedDict() # 选择区域 self.rawImageEditor = None self.currentListBoxSelIdx = None #当前选择的项目 self.setupUI() def scaleDisplayImage(self, image) : w = image.width h = image.height self.imageScale = 1.0 # 计算最小的缩放比例,保证原始宽高比 if w > self.imageDispWidth and h > self.imageDispHeight : ws = self.imageDispWidth * 1.0 / w hs = self.imageDispHeight * 1.0 / h if ws < hs : self.imageScale = ws else : self.imageScale = hs elif w > self.imageDispWidth and h < self.imageDispHeight : self.imageScale = self.imageDispWidth * 1.0 / w elif w < self.imageDispWidth and h > self.imageDispHeight : self.imageScale = self.imageDispHeight * 1.0 / h # 图片缩放 return image.resize((int(self.imageScale * w), int(self.imageScale * h))) def loadImageCfgFile(self, imf): (path, name) = os.path.split(imf) cfgname = imf + '.txt' if (not os.path.exists(cfgname)) or (not os.path.isfile(cfgname)) : cfgname = os.path.join(path, 'mask', name) + '.txt' if (not os.path.exists(cfgname)) or (not os.path.isfile(cfgname)) : cfgname = os.path.join(path, 'cfg', name) + '.txt' self.liRect.clear() if os.path.exists(cfgname) and os.path.isfile(cfgname) : with open(cfgname, "r") as f: lines = f.readlines() for line in lines: rs = line.split(',') r = (float(rs[0].strip()), float(rs[1].strip()), float(rs[2].strip()), float(rs[3].strip())) if len(rs) > 4 : self.liRect[r] = rs[4].strip() else : self.liRect[r] = '' # 打开图片时使用,传值(图)给展示函数 def openAndDisplayImage(self): imF = self.selectImageFile() if '' != imF : self.rawImagePath = imF self.loadImageCfgFile(self.rawImagePath) self.drawListBox() self.rawImage = Image.open(self.rawImagePath) self.rawImage = self.rawImage.convert('RGBA') self.drawRawImageDisp() self.image_l_trans.image = None self.transRawImage = None def drawListBox(self): self.l_box.delete(0, tk.END) for r in self.liRect.keys(): r = '{},{},{},{}'.format(round(r[0],1), round(r[1],1), round(r[2],1), round(r[3],1)) self.l_box.insert(0, r) def drawRawImageDisp(self, selItems=[]): self.dispImage = self.scaleDisplayImage(self.rawImage) self.dispImage = self.dispImage.convert('RGB') draw = ImageDraw.Draw(self.dispImage) rs = list(self.liRect.keys()) for i in range(len(rs)) : r = rs[i] if i in selItems : draw.rectangle((r[0] * self.imageScale, r[1] * self.imageScale, r[2] * self.imageScale, r[3] * self.imageScale), outline = "red") else : draw.rectangle((r[0] * self.imageScale, r[1] * self.imageScale, r[2] * self.imageScale, r[3] * self.imageScale), outline = "green") img = ImageTk.PhotoImage(self.dispImage) self.image_l_raw.config(image=img) self.image_l_raw.image = img def deleteSelectedItemFromListBox(self): #print(self.l_box.get(self.l_box.curselection())) idx = self.l_box.curselection() if len(idx) > 0 and (None == self.rawImageEditor) : ro = collections.OrderedDict() rs = self.liRect.keys() for i in range(len(rs)) : if i not in idx : r = rs[i] ro[r] = self.liRect[r] self.liRect = ro self.drawListBox() self.drawRawImageDisp() # 打开图片时使用,获得地址 def selectImageFile(self): path = tk.StringVar() file_entry = tk.Entry(self, state='readonly', text=path) path_ = askopenfilename() path.set(path_) return file_entry.get() def rawImageLabelClicked(self, event): if (None != self.rawImage) and (None == self.rawImageEditor) : self.rawImageEditor = RawImageEditor(self, self.rawImage, self.liRect.keys()) self.rawImageEditor.waitForWindow(self.image_l_raw) rs = self.rawImageEditor.selectedPositions() trs = collections.OrderedDict() for k in rs : if k in self.liRect : trs[k] = self.liRect[k] else: trs[k] = '' self.liRect = trs self.rawImageEditor = None self.drawListBox() self.drawRawImageDisp() def onRectListboxSelect(self, event): idx = self.l_box.curselection() if len(idx) > 0 and (self.currentListBoxSelIdx != idx) : self.currentListBoxSelIdx = idx ctx = '' rs = list(self.liRect.keys()) for i in range(len(self.currentListBoxSelIdx)) : v = self.currentListBoxSelIdx[i] k = rs[v] ctx = ctx + self.liRect[k] if i < len(self.currentListBoxSelIdx)-1 : ctx = ctx + ',' self.edtListBoxSel.delete(0, tk.END) self.edtListBoxSel.insert(0, ctx) self.drawRawImageDisp(idx) def onEdtListBoxComplete(self, event): idx = self.l_box.curselection() if (len(idx) > 0) and (self.currentListBoxSelIdx == idx) : ctx = self.edtListBoxSel.get().strip() ctx = ctx.split(',') while len(idx) > len(ctx) : ctx.append('') rs = list(self.liRect.keys()) n = 0 for i in idx : r = rs[i] self.liRect[r] = ctx[n] n = n + 1 def drawTransImageDisp(self): transImage = self.scaleDisplayImage(self.transRawImage) transImage = transImage.convert('L') img = ImageTk.PhotoImage(transImage) self.image_l_trans.config(image=img) self.image_l_trans.image = img def doTransRawImage(self): self.transRawImage = Image.new('L', (self.rawImage.width, self.rawImage.height)) rs = self.liRect.keys() for r in rs : im = self.rawImage.crop(r) cv_im = cv.cvtColor(np.asarray(im), cv.COLOR_RGB2BGR) hsv = cv.cvtColor(cv_im, cv.COLOR_BGR2HSV) _, _, v = cv.split(hsv) avg = np.average(v.flatten()) pixels = im.load() scale = float(self.brightnessScale.get()) for j in range(im.height) : for i in range(im.width) : hv = v[j,i] if hv < avg * scale: #im.putpixel((i, j), 0) # pixels[i, j] = 0 '''else : im.putpixel((i, j), (255, 255, 255, 255))''' self.transRawImage.paste(im, (int(r[0]),int(r[1])), mask = None) self.drawTransImageDisp() def onTransRawImageBtnClicked(self): if None != self.rawImage : self.doTransRawImage() def saveTransCfg(self, pth): (path,name) = os.path.split(self.rawImagePath) cfg = os.path.join(pth, 'cfg') if os.path.exists(cfg) and os.path.isdir(cfg) : cfgname = os.path.join(cfg, name) + '.txt' else : cfgname = os.path.join(pth, name) + '.txt' with open(cfgname, "w") as f: for r in self.liRect.keys() : v = self.liRect[r] line = '{},{},{},{},{}\n'.format(r[0], r[1], r[2], r[3], v) f.write(line) def saveCombinedImage(self, pth): im_A = cv.cvtColor(np.array(self.rawImage), cv.COLOR_RGB2BGR) im_B = cv.cvtColor(np.array(self.transRawImage), cv.COLOR_RGB2BGR) im_AB = np.concatenate([im_A, im_B], 1) ext = os.path.splitext(self.rawImagePath)[-1] ext = ext.lower() name = os.path.basename(self.rawImagePath) sp = name.split('.')[:-1] comb = os.path.join(pth, 'comb') if os.path.exists(comb) and os.path.isdir(comb) : name = sp[0] + ext pth = comb else : name = sp[0] + '_comb' + ext path_AB = os.path.join(pth, name) cv.imwrite(path_AB, im_AB) def saveTransImage(self, path): ext = os.path.splitext(self.rawImagePath)[-1] ext = ext.lower() (path,name) = os.path.split(self.rawImagePath) mask = os.path.join(path, 'mask') fn = name.split('.')[:-1] if os.path.exists(mask) and os.path.isdir(mask) : fn = fn[0] + ext else : fn = fn[0] + '_mask' + ext im = os.path.join(mask, fn) if not im.endswith(ext) : im = im + ext self.transRawImage.save(im) def saveTransInformation(self, path): if '' != path : self.saveTransImage(path) self.saveTransCfg(path) self.saveCombinedImage(path) def onSaveTransRawImageBtnClicked(self): if None != self.transRawImage : (path,name) = os.path.split(self.rawImagePath) dirname = askdirectory(initialdir = os.path.expanduser(path), title = '保存结果') if '' != dirname : self.defSavePath = dirname self.saveTransInformation(self.defSavePath) def onDefaultSaveTransRawImageBtnClicked(self) : if (None != self.transRawImage) and ('' != self.defSavePath) : self.saveTransInformation(self.defSavePath) def setupUI(self): # 左边菜单栏 left_f = tk.Frame(self, height=self.frameDispHeight, width=self.leftFrameWidth) left_f.pack(side=tk.LEFT) # 各种功能按钮名称及位置 btnOpen = tk.Button(left_f, text='打开图像', command=self.openAndDisplayImage) btnOpen.place(y=25, x=30, width=300, height=self.btnHeight) btnTrans = tk.Button(left_f, text='处理图像', command=self.onTransRawImageBtnClicked) btnTrans.place(y=75, x=30, width=300, height=self.btnHeight) l_selRect = tk.Label(left_f, text = '鼠标选定区域') l_selRect.place(x=0, y=125, width=self.leftFrameWidth, height=self.labelTextHeight) '''列表''' self.l_box = tk.Listbox(left_f, exportselection=False) # 创建两个列表组件 self.l_box.place(x=1, y=125+self.labelTextHeight, width=self.leftFrameWidth-2, height=270) self.l_box.bind('<<ListboxSelect>>', self.onRectListboxSelect) self.drawListBox() # 选中列表注释 ledt = tk.Label(left_f, text = '选中区域内容注释(回车结束):', anchor = 'w') ledt.place(x=1, y=420, width=self.leftFrameWidth-2) self.edtListBoxSel = tk.Entry(left_f) self.edtListBoxSel.place(x=1, y=440, width=self.leftFrameWidth-2) self.edtListBoxSel.bind('<Return>', self.onEdtListBoxComplete) # 亮度比例调节 tkScale = tk.Scale(left_f, from_=0.2, to=1.8, orient=tk.HORIZONTAL, resolution=0.1, length=self.leftFrameWidth-2, variable=self.brightnessScale, label = '亮度比:') tkScale.place(x=1, y=480) # 删除选定项 btnDel = tk.Button(left_f, text='删除选定项', command=self.deleteSelectedItemFromListBox) btnDel.place(y=550, x=30, width=300, height=self.btnHeight) btnSave = tk.Button(left_f, text='保存结果', command=self.onSaveTransRawImageBtnClicked) btnSave.place(y=600, x=30, width=300, height=self.btnHeight) btnDefSave = tk.Button(left_f, text='默认或上次位置保存', command=self.onDefaultSaveTransRawImageBtnClicked) btnDefSave.place(y=650, x=30, width=300, height=self.btnHeight) # 右侧图像显示栏 right_f = tk.Frame(self, height=self.frameDispHeight, width=self.imageDispWidth) right_f.pack(side=tk.RIGHT) l_rawT = tk.Label(right_f, text = '原始图片') l_rawT.place(x=0, y=0, width=self.imageDispWidth, height=self.labelTextHeight) self.image_l_raw = tk.Label(right_f, relief='ridge') self.image_l_raw.place(x=0, y=self.labelTextHeight, width=self.imageDispWidth, height=self.imageDispHeight) self.image_l_raw.bind("<Button-1>",self.rawImageLabelClicked) l_transT = tk.Label(right_f, text = '处理后图片') l_transT.place(x=0, y=self.labelTextHeight + self.imageDispHeight, width=self.imageDispWidth, height=self.labelTextHeight) self.image_l_trans = tk.Label(right_f, relief='ridge') self.image_l_trans.place(x=0, y=self.labelTextHeight + self.imageDispHeight + self.labelTextHeight, width=self.imageDispWidth, height=self.imageDispHeight) if __name__ == '__main__' : win = MainWin() # 进入消息循环 win.mainloop() |
1. 参照 pytorch 1.0.1在ubuntu 18.04(GeForce GTX 760)编译(CUDA-10.1) 建立 pytorch 1.0.1
的编译环境,并解决编译时遇到的问题。
2. 依旧是推荐在 Anaconda 上建立独立的编译环境,然后执行编译:
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 |
$ sudo apt-get install git # conda remove -n Pix2Pix --all $ conda create -n Pix2Pix -y python=3.6.8 pip $ source activate Pix2Pix $ conda install numpy pyyaml mkl=2019.1 mkl-include=2019.1 setuptools cmake cffi typing pybind11 $ conda install ninja # magma-cuda90 magma-cuda91 magma-cuda92 会编译失败 $ conda install -c pytorch magma-cuda101 $ git clone https://github.com/pytorch/pytorch $ cd pytorch # pytorch 1.0.1 版本支持“Compute Capability” 低于3.0版本的硬件,pytorch 1.2.0需要至少3.5版本的硬件才可以正常运行 # https://github.com/pytorch/pytorch/blob/v1.3.0/torch/utils/cpp_extension.py $ git checkout v1.0.1 -b v1.0.1 $ git submodule sync $ git submodule update --init --recursive $ export CMAKE_PREFIX_PATH=${CONDA_PREFIX:-"$(dirname $(which conda))/../"} # 如果不需要使用cuda的话,这里还要加上一句:export NO_CUDA=1 $ python setup.py clean # 卸载以前安装的pytorch $ conda uninstall pytorch # 从Nvidia开发网站查询到自己硬件对应的“Compute Capability” # 比如 “GeForce GTX 760” 对应 “3.0” 计算能力,能力不正确会导致运行异常 # RuntimeError: cuda runtime error (48) : no kernel image is available for execution on the device $ python setup.py install # 对于开发者模式,可以使用 # python setup.py build develop # 一定要退出 pytorch 的编译目录,在pytorch代码目录下执行命令会出现异常 $ cd .. # 退出环境 $ conda deactivate |
编译出错信息,参考 pytorch 1.0.1在ubuntu 18.04(GeForce GTX 760)编译(CUDA-10.1) 里面的介绍解决。
3. 编译安装 TorchVision
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 |
$ sudo apt-get install git # 进入运行环境 $ source activate Pix2Pix $ git clone https://github.com/pytorch/vision.git # 也可本站下载一份拷贝 wget https://www.mobibrw.com/wp-content/uploads/2019/11/vision.zip $ cd vision $ git checkout v0.2.1 -b v0.2.1 $ python setup.py install # 退出环境 $ conda deactivate |
4. 检出 CycleGAN and pix2pix in PyTorch 的代码,并安装依赖
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 |
# 进入运行环境 $ source activate Pix2Pix $ git clone https://github.com/junyanz/pytorch-CycleGAN-and-pix2pix.git # 也可本站下载 wget https://www.mobibrw.com/wp-content/uploads/2019/12/pytorch-CycleGAN-and-pix2pix.zip $ cd pytorch-CycleGAN-and-pix2pix # 下载人脸替换部分的数据集 $ bash datasets/download_pix2pix_dataset.sh facades # 也可本站下载然后自己参照脚本解压缩到指定目录 https://www.mobibrw.com/wp-content/uploads/2019/12/facades.tar.gz # 安装依赖 $ pip install pillow==6.2.1 $ pip install dominate==2.4.0 $ pip install visdom==0.1.8.9 # 修正错误 models/networks.py # TypeError: cuda() got an unexpected keyword argument 'device_id' $ sed -i "s/netG\.cuda(device_id=gpu_ids\[0\])/netG.cuda(gpu_ids[0])/g" models/networks.py $ sed -i "s/netD\.cuda(device_id=gpu_ids\[0\])/netD.cuda(gpu_ids[0])/g" models/networks.py $ sed -i "s/network\.cuda(device_id=gpu_ids\[0\])/network.cuda(gpu_ids[0])/g" models/base_model.py # 开启WEB服务,主要是第一次运行需要下载部分辅助软件包, # 训练之前需要执行,否则下面训练的时候会报错 $ python -m visdom.server & # 等待屏幕上出现 “You can navigate to http://localhost:8097” 代表服务启动成功 # 执行训练 $ bash scripts/train_pix2pix.sh |
执行训练的时候,如果出现如下错误:
1 2 3 4 5 6 |
Traceback (most recent call last): File "train.py", line 47, in <module> errors = model.get_current_errors() File "~/pytorch-CycleGAN-and-pix2pix/models/pix2pix_model.py", line 122, in get_current_errors return OrderedDict([('G_GAN', self.loss_G_GAN.data[0]), IndexError: invalid index of a 0-dim tensor. Use tensor.item() to convert a 0-dim tensor to a Python number |
这个原因是由于 PyTorch 版本差异造成的,(作者在 Pytorch 0.4.1
版本上测试,我们在 Pytorch 1.0.1
版本上测试),执行如下命令修复:
1 2 3 4 5 6 7 8 9 |
#loss_G_GAN.data[0] 替换为 loss_G_GAN.item() $ sed -i "s/self\.loss_G_GAN\.data\[0]/self.loss_G_GAN.item()/g" models/pix2pix_model.py $ sed -i "s/self\.loss_G_L1\.data\[0]/self.loss_G_L1.item()/g" models/pix2pix_model.py $ sed -i "s/self\.loss_D_real\.data\[0]/self.loss_D_real.item()/g" models/pix2pix_model.py $ sed -i "s/self\.loss_D_fake\.data\[0]/self.loss_D_fake.item()/g" models/pix2pix_model.py |
5. 测试训练结果
1 2 3 |
$ bash scripts/test_pix2pix.sh # 观察结果需要打开 ./results/facades_pix2pix/test_latest/index.html |