load old nets

Author: culurciello

.gitignore

@@ -1,2 +1,4 @@
 
 *.ipynb
+
+*.pyc

visor/visor.py

@@ -39,14 +39,6 @@ def cat_file():
             quit()
     return categories
 
-# image pre-processing functions:
-transformsImage = transforms.Compose([
-        # transforms.ToPILImage(),
-        # transforms.Scale(256),
-        # transforms.CenterCrop(224),
-        transforms.ToTensor(),
-        transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225]) # needed for pythorch ZOO models on ImageNet (stats)
-    ])
 
 print("Visor demo e-Lab")
 xres = 640
@@ -66,6 +58,15 @@ def cat_file():
 # print model
 softMax = nn.Softmax() # to get probabilities out of CNN
 
+# image pre-processing functions:
+transformsImage = transforms.Compose([
+        # transforms.ToPILImage(),
+        # transforms.Scale(256),
+        # transforms.CenterCrop(224),
+        transforms.ToTensor(),
+        transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225]) # needed for pythorch ZOO models on ImageNet (stats)
+    ])
+
 while True:
     startt = time.time()
     ret, frame = cam.read()

visor/visor_oldnets.py

@@ -0,0 +1,143 @@
+#! /usr/local/bin/python3
+
+import sys
+import os
+import time
+import cv2 # install cv3, python3:  http://seeb0h.github.io/howto/howto-install-homebrew-python-opencv-osx-el-capitan/
+# add to profile: export PYTHONPATH=$PYTHONPATH:/usr/local/Cellar/opencv3/3.2.0/lib/python3.6/site-packages/
+import numpy as np
+import argparse
+import torch
+import torchvision.transforms as transforms
+from torch.utils.serialization import load_lua
+from torch.legacy import nn # import torch.nn as nn
+
+
+def define_and_parse_args():
+    # argument Checking
+    parser = argparse.ArgumentParser(description="Visor Demo")
+    # parser.add_argument('network', help='CNN model file')
+    parser.add_argument('categories', help='text file with categories')
+    parser.add_argument('-i', '--input',  type=int, default='0', help='camera device index or file name, default 0')
+    parser.add_argument('-s', '--size', type=int, default=224, help='network input size')
+    # parser.add_argument('-S', '--stat', help='stat.txt file')
+    return parser.parse_args()
+
+
+def cat_file():
+    # load classes file
+    categories = []
+    if hasattr(args, 'categories') and args.categories:
+        try:
+            f = open(args.categories, 'r')
+            for line in f:
+                cat = line.split(',')[0].split('\n')[0]
+                if cat != 'classes':
+                    categories.append(cat)
+            f.close()
+            print('Number of categories:', len(categories))
+        except:
+            print('Error opening file ' + args.categories)
+            quit()
+    return categories
+
+
+print("Visor demo e-Lab - older Torch7 networks")
+xres = 640
+yres = 480
+args = define_and_parse_args()
+categories = cat_file() # load category file
+print(categories)
+
+
+# setup camera input:
+cam = cv2.VideoCapture(args.input)
+cam.set(3, xres)
+cam.set(4, yres)
+
+# load old-pre-trained Torch7 CNN model:
+
+# https://www.dropbox.com/sh/l0rurgbx4k6j2a3/AAA223WOrRRjpe9bzO8ecpEpa?dl=0
+model = load_lua('/Users/eugenioculurciello/Dropbox/shared/models/elab-alexowt-46/model.net')
+stat = load_lua('/Users/eugenioculurciello/Dropbox/shared/models/elab-alexowt-46/stat.t7')
+model.modules[13] = nn.View(1,9216)
+
+# https://www.dropbox.com/sh/xcm8xul3udwo72o/AAC8RChVSOmgN61nQ0cyfdava?dl=0
+# model = load_lua('/Users/eugenioculurciello/Dropbox/shared/models/elab-alextiny-46/model.net')
+# stat = load_lua('/Users/eugenioculurciello/Dropbox/shared/models/elab-alextiny-46/stat.t7')
+# model.modules[13] = nn.View(1,64)
+
+# https://www.dropbox.com/sh/anklohs9g49z1o4/AAChA9rl0FEGixT75eT38Dqra?dl=0
+# model = load_lua('/Users/eugenioculurciello/Dropbox/shared/models/elab-enet-demo-46/model.net')
+# stat = load_lua('/Users/eugenioculurciello/Dropbox/shared/models/elab-enet-demo-46/stat.t7')
+# model.modules[41] = nn.View(1,1024)
+
+# https://www.dropbox.com/sh/s0hwugnmhwkk9ow/AAD_abZ2LOav9GeMETt5VGvGa?dl=0
+# model = load_lua('/Users/eugenioculurciello/Dropbox/shared/models/enet128-demo-46/model.net')
+# stat = load_lua('/Users/eugenioculurciello/Dropbox/shared/models/enet128-demo-46/stat.t7')
+# model.modules[32] = nn.View(1,512)
+
+# print(model)
+# this now should work:
+# model.forward(torch.Tensor(1,3,224,224)) # test
+
+# image pre-processing functions:
+transformsImage = transforms.Compose([
+        # transforms.ToPILImage(),
+        # transforms.Scale(256),
+        # transforms.CenterCrop(224),
+        transforms.ToTensor(),
+        transforms.Normalize(stat.mean, stat.std)
+    ])
+
+while True:
+    startt = time.time()
+    ret, frame = cam.read()
+    if not ret:
+        break
+
+    if xres > yres:
+        frame = frame[:,int((xres - yres)/2):int((xres+yres)/2),:]
+    else:
+        frame = frame[int((yres - xres)/2):int((yres+xres)/2),:,:]
+    
+    pframe = cv2.resize(frame, dsize=(args.size, args.size))
+    
+    # prepare and normalize frame for processing:
+    pframe = np.swapaxes(pframe, 0, 2)
+    pframe = np.expand_dims(pframe, axis=0)
+    pframe = transformsImage(pframe)
+    
+    # process via CNN model:
+    output = model.forward(pframe)
+    if output is None:
+        print('no output from CNN model file')
+        break
+
+    # print(output)
+    output = output.numpy()[0]
+
+    # process output and print results:
+    order = output.argsort()
+    last = len(categories)-1
+    text = ''
+    for i in range(min(5, last+1)):
+      text += categories[order[last-i]] + ' (' + '{0:.2f}'.format(output[order[last-i]]*100) + '%) '
+
+    # overlay on GUI frame
+    # cv2.displayOverlay('win', text, 1000) # if Qt is present!
+    font = cv2.FONT_HERSHEY_SIMPLEX
+    cv2.putText(frame, text, (10, yres-20), font, 0.5, (255, 255, 255), 1)
+    cv2.imshow('win', frame)
+
+    endt = time.time()
+    # sys.stdout.write("\r"+text+"fps: "+'{0:.2f}'.format(1/(endt-startt))) # text output 
+    sys.stdout.write("\rfps: "+'{0:.2f}'.format(1/(endt-startt)))
+    sys.stdout.flush()
+    
+    if cv2.waitKey(1) == 27: # ESC to stop
+        break
+
+# end program:
+cam.release()
+cv2.destroyAllWindows()