migrate to PIL for image operations

README.md

@@ -91,7 +91,9 @@ python test.py --dataset university --detection_threshold 0.99 --viz
 3. matplotlib
 4. h5py
 5. tensorflow
-6. [Deeplab V3](https://github.com/sthalles/deeplab_v3)
+6. PIL
+7. scikit-learn
+8. [Deeplab V3](https://github.com/sthalles/deeplab_v3)
 
 ## Examples
 

annotate.py

@@ -1,8 +1,9 @@
 import matplotlib.pyplot as plt
 import numpy
-import scipy.misc
+from PIL import Image
 from sklearn.cluster import KMeans
 import sys
+import os
 
 dataset = 'beach'
 margin = 10
@@ -18,13 +19,14 @@
 
 def next_img():
 	global image_np, image_display, labels, img_id, segment_id
-	try:
-		image_np = scipy.misc.imread('dataset/%s/%d.png'%(dataset,img_id))
-	except IOError:
+	image_filename = 'dataset/%s/%d.png'%(dataset,img_id)
+	if os.path.exists(image_filename):
+		image_np = numpy.array(Image.open(image_filename))
+	else:
 		sys.exit(0)
 	image_display = image_np.copy()
 	image_np = numpy.mean(image_np, axis=2)
-	#print(image_np.shape, image_display.shape)
+#	print(image_np.shape, image_display.shape)
 	labels = numpy.zeros(image_np.shape, dtype=int)
 	segment_id=0
 	plt.imshow(image_display)
@@ -35,7 +37,7 @@ def next_img():
 def onkey(event):
 	if event.key==' ':
 		global img_id
-		scipy.misc.imsave('dataset/%s/label%d.png'%(dataset,img_id), labels)
+		Image.fromarray(labels, mode='L').save('dataset/%s/label%d.png'%(dataset,img_id))
 		img_id += 1
 		next_img()
 	elif event.key=='r':

baselines.py

@@ -1,12 +1,13 @@
 import matplotlib.pyplot as plt
 import numpy
-import scipy.misc
+from PIL import Image
 import sys
 import cv2
 import time
 from sklearn.cluster import KMeans
 from sklearn.mixture import GaussianMixture
 import glob
+import os
 
 method = 'threshold'
 #method = 'threshold_adp'
@@ -44,13 +45,14 @@
 	if method!='backSub' and image_id < test_idx:
 		image_id += 1
 		continue
-	try:
-		I = scipy.misc.imread('dataset/%s/%d.png' % (dataset,image_id))
+	image_filename = 'dataset/%s/%d.png' % (dataset,image_id)
+	if os.path.exists(image_filename):
+		I = numpy.array(Image.open(image_filename))
 		if len(I.shape)>2:
 			I = numpy.mean(I, axis=2)
-	except IOError:
+	else:
 		break
-	gt = scipy.misc.imread('dataset/%s/label%d.png' % (dataset, image_id))
+	gt = numpy.array(Image.open('dataset/%s/label%d.png'%(dataset,image_id)))
 	gt = gt > 0
 	dt = numpy.zeros(I.shape, dtype=bool)
 	image_np = I[ybound:ybound+imscale, xbound:xbound+imscale]
@@ -71,7 +73,7 @@
 		if dataset=='combined' and image_id in [97, 225, 249]:
 			backSub = cv2.createBackgroundSubtractorMOG2()
 		mask = backSub.apply(image_np)
-		scipy.misc.imsave('dataset/%s/backSub/%d.png'%(dataset,image_id), mask.astype(numpy.uint8))
+		Image.fromarray(mask.astype(numpy.uint8), mode='L').save('dataset/%s/backSub/%d.png'%(dataset,image_id))
 	elif method=='kmeans':
 		window_size = 15 if dataset=='beach' or dataset=='shore' else 100
 		margin = 10 if dataset=='beach' or dataset=='shore' else 100
@@ -227,9 +229,9 @@
 #	print('Image #%d Precision:%.2f/%.2f Recall:%.2f/%.2f (%.2fs)'%(image_id, prc,obj_prc,rcl,obj_rcl, t2-t1))
 
 	if image_id == save_frame:
-		scipy.misc.imsave('results/original_%d.png'%save_frame, image_np.astype(numpy.uint8))
-		scipy.misc.imsave('results/detected_%s_%d.png'%(method,save_frame), dt_viz)
-		scipy.misc.imsave('results/ground_truth_%d.png'%save_frame, gt_viz)
+		Image.fromarray(image_np.astype(numpy.uint8), mode='L').save('results/original_%d.png'%save_frame)
+		Image.fromarray(dt_viz, mode='RGB').save('results/detected_%s_%d.png'%(method, save_frame))
+		Image.fromarray(gt_viz, mode='RGB').save('results/ground_truth_%d.png'%save_frame)
 		print('save_frame',save_frame)
 		sys.exit(1)
 	if viz:

process_record.py

@@ -1,5 +1,5 @@
 import numpy
-import scipy.misc
+from PIL import Image
 import h5py
 import glob
 import sys
@@ -34,29 +34,29 @@
 previous_img = None
 
 for i in range(1,num_samples+1):
-	image_np = scipy.misc.imread('dataset/%s/%d.png'%(dataset,i),mode='RGB')
+	image_np = numpy.array(Image.open('dataset/%s/%d.png'%(dataset,i)))
 	image_np = image_np[ybound:ybound+imscale,xbound:xbound+imscale].mean(axis=2)
 	if previous_img is None:
 		diff_img = numpy.zeros(image_np.shape, dtype=numpy.uint8)
 		backSub_img = numpy.zeros(image_np.shape, dtype=numpy.uint8)
 	else:
 		diff_img = ((image_np - previous_img)/2 + 128).astype(numpy.uint8)
 		if use_history:
-			backSub_img = scipy.misc.imread('dataset/%s/backSub/%d.png'%(dataset,i),mode='RGB').mean(axis=2)
+			backSub_img = numpy.array(Image.open('dataset/%s/backSub/%d.png'%(dataset,i))).mean(axis=2)
 	previous_img = image_np
 	image_h = image_np.shape[0]
 	image_w = image_np.shape[1]
 	if use_history:
 		image_np = numpy.dstack((image_np, diff_img, backSub_img)).astype(numpy.uint8)
 	else:
 		image_np = numpy.dstack((image_np, image_np, image_np)).astype(numpy.uint8)
-	annotation = scipy.misc.imread('dataset/%s/label%d.png'%(dataset,i), mode='L')
+	annotation = numpy.array(Image.open('dataset/%s/label%d.png'%(dataset,i)))
 	for p in numpy.array(numpy.nonzero(annotation)).T:
 		pos_idx.add(tuple(p))
 	annotation = annotation[ybound:ybound+imscale,xbound:xbound+imscale]
 	if imscale!=imsize:
-		image_np = scipy.misc.imresize(image_np, size=(imsize, imsize), interp='bilinear')
-		annotation = scipy.misc.imresize(annotation, size=(imsize, imsize), interp='bilinear')
+		image_np = numpy.array(Image.fromarray(image_np).resize((imsize, imsize), resample=Image.BILINEAR))
+		annotation = numpy.array(Image.fromarray(annotation).resize((imsize, imsize), resample=Image.BILINEAR))
 	annotation = numpy.array(annotation > 0, dtype=numpy.uint8)
 	print(i,image_np.shape,image_np.dtype)
 	if i-1 in train_samples:

test.py

@@ -13,6 +13,7 @@
 import sys
 import sklearn.metrics
 import glob
+from PIL import Image
 
 parser = argparse.ArgumentParser()
 envarg = parser.add_argument_group('Training params')
@@ -108,28 +109,28 @@ def __init__(self):
 previous_img = None
 for image_idx in range(1,num_samples+1):
 	t1 = time.time()
-	original_image = scipy.misc.imread('dataset/%s/%d.png'%(args.dataset,image_idx),mode='RGB')
+	original_image = numpy.array(Image.open('dataset/%s/%d.png'%(args.dataset,image_idx)))
 	image_np = original_image[ybound:ybound+imscale,xbound:xbound+imscale].mean(axis=2)
 	if args.use_history:
 		if previous_img is None:
 			diff_img = numpy.zeros(image_np.shape, dtype=numpy.uint8)
 			backSub_img = numpy.zeros(image_np.shape, dtype=numpy.uint8)
 		else:
 			diff_img = ((image_np - previous_img)/2 + 128).astype(numpy.uint8)
-			backSub_img = scipy.misc.imread('dataset/%s/backSub/%d.png'%(args.dataset,image_idx),mode='RGB').mean(axis=2)
+			backSub_img = numpy.array(Image.open('dataset/%s/backSub/%d.png'%(args.dataset,image_idx))).mean(axis=2)
 		previous_img = image_np
 	if image_idx < test_idx:
 		continue
 	if args.use_history:
 		image_np = numpy.dstack((image_np, diff_img, backSub_img)).astype(numpy.uint8)
 	else:
 		image_np = numpy.dstack((image_np, image_np, image_np)).astype(numpy.uint8)
-	original_annotation = scipy.misc.imread('dataset/%s/label%d.png'%(args.dataset,image_idx), mode='L')
+	original_annotation = numpy.array(Image.open('dataset/%s/label%d.png'%(args.dataset,image_idx)))
 	annotation = original_annotation[ybound:ybound+imscale,xbound:xbound+imscale]
 	annotation = numpy.array(annotation > 0, dtype=numpy.uint8)
 	if imsize!=imscale:
-		input_images[:] = scipy.misc.imresize(image_np, size=(imsize, imsize), interp='bilinear')
-		input_labels[:] = scipy.misc.imresize(annotation, size=(imsize, imsize), interp='bilinear')
+		input_images[:] = numpy.array(Image.fromarray(image_np).resize((imsize, imsize), resample=Image.BILINEAR))
+		input_labels[:] = numpy.array(Image.fromarray(annotation).resize((imsize, imsize), resample=Image.BILINEAR))
 	else:
 		input_images[:] = image_np
 		input_labels[:] = annotation
@@ -142,7 +143,7 @@ def __init__(self):
 #	print('Loss %.2f(%.2f+%.2f) tp:%d fp:%d fn:%d prc:%.2f rcl:%.2f'%(ls,pl,nl,vtp,vfp,vfn,prc,rcl))
 	predicted_softmax = scipy.special.softmax(result[0,:,:,:],axis=-1)[:,:,1]
 	if imsize!=imscale:
-		predicted_softmax = scipy.misc.imresize(predicted_softmax, size=(imscale, imscale), interp='bilinear') / 255.0
+		predicted_softmax = numpy.array(Image.fromarray(predicted_softmax, mode='F').resize((imscale, imscale), resample=Image.BILINEAR))
 	ap_dt.extend(predicted_softmax.flatten())
 	ap_gt.extend(annotation.flatten())
 	predicted = predicted_softmax > detection_threshold
@@ -222,12 +223,12 @@ def __init__(self):
 		dt_viz[y1:y2, x2, :] = [255,255,0]
 
 	comp_time.append(t2 - t1)
-#	print('%d/%d images prc %.2f/%.2f rcl %.2f/%.2f (%.2fs)'%(image_idx,num_samples,prc,obj_prc,rcl,obj_rcl,t2-t1))
+	print('%d/%d images prc %.2f/%.2f rcl %.2f/%.2f (%.2fs)'%(image_idx,num_samples,prc,obj_prc,rcl,obj_rcl,t2-t1))
 
 	if image_idx == args.save_frame:
-		scipy.misc.imsave('results/original_%d.png'%args.save_frame, image_np[:,:,0].astype(numpy.uint8))
-		scipy.misc.imsave('results/detected_%s_%d.png'%('history' if args.use_history else 'cnn', args.save_frame), dt_viz)
-		scipy.misc.imsave('results/ground_truth_%d.png'%args.save_frame, gt_viz)
+		Image.fromarray(image_np[:,:,0].astype(numpy.uint8), mode='L').save('results/original_%d.png'%args.save_frame)
+		Image.fromarray(dt_viz, mode='RGB').save('results/detected_%s_%d.png'%('history' if args.use_history else 'cnn', args.save_frame))
+		Image.fromarray(gt_viz, mode='RGB').save('results/ground_truth_%d.png'%args.save_frame)
 		print('save_frame',args.save_frame)
 		sys.exit(1)
 	if viz: