Decoupling model creation from training script. Use of threads for da…

…ta generation. Small changes in the run script.

README.md

@@ -30,7 +30,7 @@ Use the script "run.sh" to run our ALPR approach. It requires 3 arguments:
 * __CSV file:__ specify an output CSV file.
 
 ```shellscript
-$ bash run.sh samples/test /tmp/output /tmp/output/results.csv
+$ bash run.sh -i samples/test -o /tmp/output -c /tmp/output/results.csv
 ```
 
 ## Training the LP detector
@@ -40,10 +40,12 @@ To train the LP detector network from scratch, or fine-tuning it for new samples
 The following command can be used to train the network from scratch considering the data inside the train-detector folder:
 
 ```shellscript
-$ python train-detector.py --name new-network --outdir /tmp/ --input-dir samples/train-detector
+$ mkdir models
+$ python create-model.py eccv models/eccv-model-scracth
+$ python train-detector.py --model models/eccv-model-scracth --name my-trained-model --train-dir samples/train-detector --output-dir models/my-trained-model/ -op Adam -lr .001 -its 300000 -bs 64
 ```
 
-For fine-tunning, add "-m data/lp-detector/wpod-net" to the command line above.
+For fine-tunning, use your model with --model option.
 
 ## A word on GPU and CPU
 

src/detector_network.py → create-model.py

@@ -1,6 +1,12 @@
 
+import sys
+import keras
+
 from keras.layers import Conv2D, MaxPooling2D, BatchNormalization, Add, Activation, Concatenate, Input
 from keras.models import Model
+from keras.applications.mobilenet import MobileNet
+
+from src.keras_utils import save_model
 
 
 def res_block(x,sz,filter_sz=3,in_conv_size=1):
@@ -21,7 +27,13 @@ def conv_batch(_input,fsz,csz,activation='relu',padding='same',strides=(1,1)):
  output = Activation(activation)(output)
  return output
 
-def create_model():
+def end_block(x):
+ xprobs = Conv2D(2, 3, activation='softmax', padding='same')(x)
+ xbbox = Conv2D(6, 3, activation='linear' , padding='same')(x)
+ return Concatenate(3)([xprobs,xbbox])
+
+
+def create_model_eccv():
 
  input_layer = Input(shape=(None,None,3),name='input')
 
@@ -45,9 +57,51 @@ def create_model():
  x = res_block(x,128)
  x = res_block(x,128)
 
- xprobs = Conv2D(2, 3, activation='softmax', padding='same')(x)
- xbbox = Conv2D(6, 3, activation='linear' , padding='same')(x)
+ x = end_block(x)
+
+ return Model(inputs=input_layer,outputs=x)
+
+
+# Model not converging...
+def create_model_mobnet():
+
+ input_layer = Input(shape=(None,None,3),name='input')
+ x = input_layer
+
+ mbnet = MobileNet(input_shape=(224,224,3),include_top=True)
+
+ backbone = keras.models.clone_model(mbnet)
+ for i,bblayer in enumerate(backbone.layers[1:74]):
+ layer = bblayer.__class__.from_config(bblayer.get_config())
+ layer.name = 'backbone_' + layer.name
+ x = layer(x)
+
+ x = end_block(x)
+
+ model = Model(inputs=input_layer,outputs=x)
+
+ backbone_layers = {'backbone_' + layer.name: layer for layer in backbone.layers}
+ for layer in model.layers:
+ if layer.name in backbone_layers:
+ print 'setting ' + layer.name
+ layer.set_weights(backbone_layers[layer.name].get_weights())
+
+ return model
+
+
+if __name__ == '__main__':
+
+ modules = [func.replace('create_model_','') for func in dir(sys.modules[__name__]) if 'create_model_' in func]
+
+ assert sys.argv[1] in modules, \
+ 'Model name must be on of the following: %s' % ', '.join(modules)
+
+ modelf = getattr(sys.modules[__name__],'create_model_' + sys.argv[1])
+
+ print 'Creating model %s' % sys.argv[1]
+ model = modelf()
+ print 'Finished'
 
- x = Concatenate(3)([xprobs,xbbox])
+ print 'Saving at %s' % sys.argv[2]
+ save_model(model,sys.argv[2])
 
- return Model(input=input_layer,output=x)

license-plate-detection.py

@@ -1,6 +1,7 @@
 import sys, os
 import keras
 import cv2
+import traceback
 
 from src.keras_utils import load_model
 from glob import glob
@@ -15,40 +16,49 @@ def adjust_pts(pts,lroi):
 
 
 if __name__ == '__main__':
-
- input_dir = sys.argv[1]
- output_dir = input_dir
 
- lp_threshold = .5
+ try:
+
+ input_dir = sys.argv[1]
+ output_dir = input_dir
 
- wpod_net_path = sys.argv[2]
- wpod_net = load_model(wpod_net_path)
+ lp_threshold = .5
 
- imgs_paths = glob('%s/*car.png' % input_dir)
+ wpod_net_path = sys.argv[2]
+ wpod_net = load_model(wpod_net_path)
 
- print 'Searching for license plates using WPOD-NET'
+  imgs_paths = glob('%s/*car.png' % input_dir)
 
- for i,img_path in enumerate(imgs_paths):
+  print 'Searching for license plates using WPOD-NET'
 
- print '\t Processing %s' % img_path
+ for i,img_path in enumerate(imgs_paths):
 
- bname = splitext(basename(img_path))[0]
- Ivehicle = cv2.imread(img_path)
+ print '\t Processing %s' % img_path
 
- ratio = float(max(Ivehicle.shape[:2]))/min(Ivehicle.shape[:2])
- side = int(ratio*288.)
- bound_dim = min(side + (side%(2**4)),608)
- print "\t\tBound dim: %d, ratio: %f" % (bound_dim,ratio)
+ bname = splitext(basename(img_path))[0]
+ Ivehicle = cv2.imread(img_path)
 
- Llp,LlpImgs,_ = detect_lp(wpod_net,im2single(Ivehicle),bound_dim,2**4,(240,80),lp_threshold)
+ ratio = float(max(Ivehicle.shape[:2]))/min(Ivehicle.shape[:2])
+ side = int(ratio*288.)
+ bound_dim = min(side + (side%(2**4)),608)
+ print "\t\tBound dim: %d, ratio: %f" % (bound_dim,ratio)
 
- if len(LlpImgs):
- Ilp = LlpImgs[0]
- Ilp = cv2.cvtColor(Ilp, cv2.COLOR_BGR2GRAY)
- Ilp = cv2.cvtColor(Ilp, cv2.COLOR_GRAY2BGR)
+ Llp,LlpImgs,_ = detect_lp(wpod_net,im2single(Ivehicle),bound_dim,2**4,(240,80),lp_threshold)
 
- s = Shape(Llp[0].pts)
+ if len(LlpImgs):
+ Ilp = LlpImgs[0]
+ Ilp = cv2.cvtColor(Ilp, cv2.COLOR_BGR2GRAY)
+ Ilp = cv2.cvtColor(Ilp, cv2.COLOR_GRAY2BGR)
+
+ s = Shape(Llp[0].pts)
+
+ cv2.imwrite('%s/%s_lp.png' % (output_dir,bname),Ilp*255.)
+ writeShapes('%s/%s_lp.txt' % (output_dir,bname),[s])
+
+ except:
+ traceback.print_exc()
+ sys.exit(1)
+
+ sys.exit(0)
 
- cv2.imwrite('%s/%s_lp.png' % (output_dir,bname),Ilp*255.)
- writeShapes('%s/%s_lp.txt' % (output_dir,bname),[s])
 

license-plate-ocr.py

@@ -1,6 +1,7 @@
 import sys
 import cv2
 import numpy as np
+import traceback
 
 import darknet.python.darknet as dn
 
@@ -12,44 +13,52 @@
 
 
 if __name__ == '__main__':
+
+ try:
 
- input_dir = sys.argv[1]
- output_dir = input_dir
+ input_dir = sys.argv[1]
+ output_dir = input_dir
+
+ ocr_threshold = .4
+
+ ocr_weights = 'data/ocr/ocr-net.weights'
+ ocr_netcfg = 'data/ocr/ocr-net.cfg'
+ ocr_dataset = 'data/ocr/ocr-net.data'
 
- ocr_threshold = .4
+ ocr_net = dn.load_net(ocr_netcfg, ocr_weights, 0)
+ ocr_meta = dn.load_meta(ocr_dataset)
 
- ocr_weights = 'data/ocr/ocr-net.weights'
- ocr_netcfg = 'data/ocr/ocr-net.cfg'
- ocr_dataset = 'data/ocr/ocr-net.data'
+ imgs_paths = sorted(glob('%s/*lp.png' % output_dir))
 
- ocr_net = dn.load_net(ocr_netcfg, ocr_weights, 0)
- ocr_meta = dn.load_meta(ocr_dataset)
+ print 'Performing OCR...'
 
- imgs_paths = sorted(glob('%s/*lp.png' % output_dir))
+  for i,img_path in enumerate(imgs_paths):
 
- print 'Performing OCR...'
+  print '\tScanning %s' % img_path
 
- for i,img_path in enumerate(imgs_paths):
+  bname = basename(splitext(img_path)[0])
 
- print '\tScanning %s' % img_path
+  R,(width,height) = detect(ocr_net, ocr_meta, img_path ,thresh=ocr_threshold, nms=None)
 
- bname = basename(splitext(img_path)[0])
+  if len(R):
 
- R,(width,height) = detect(ocr_net, ocr_meta, img_path ,thresh=ocr_threshold, nms=None)
+ L = dknet_label_conversion(R,width,height)
+ L = nms(L,.45)
 
- if len(R):
+ L.sort(key=lambda x: x.tl()[0])
+ lp_str = ''.join([chr(l.cl()) for l in L])
 
- L = dknet_label_conversion(R,width,height)
- L = nms(L,.45)
+  with open('%s/%s_str.txt' % (output_dir,bname),'w') as f:
+  f.write(lp_str + '\n')
 
- L.sort(key=lambda x: x.tl()[0])
- lp_str = ''.join([chr(l.cl()) for l in L])
+ print '\t\tLP: %s' % lp_str
 
- with open('%s/%s_str.txt' % (output_dir,bname),'w') as f:
- f.write(lp_str + '\n')
+ else:
 
- print '\t\tLP: %s' % lp_str
+  print 'No characters found'
 
- else:
+ except:
+ traceback.print_exc()
+ sys.exit(1)
 
-  print 'No characters found'
+ sys.exit(0)

run.sh

@@ -27,61 +27,80 @@ retval=$?
 if [ $retval -eq 0 ]
 then
  echo "Darknet is not compiled! Go to 'darknet' directory and 'make'!"
- exit 0
+ exit 1
 fi
 
+lp_model="data/lp-detector/wpod-net.h5"
+input_dir=''
+output_dir=''
+csv_file=''
+
+
 # Check # of arguments
-if [ ! $# -eq 3 ]
-then
+usage() {
+ echo ""
+ echo " Usage:"
  echo ""
- echo " Required arguments:"
+ echo "  bash $0 -i input/dir -o output/dir -c csv_file.csv [-h] [-l path/to/model]:"
  echo ""
- echo " 1. Input dir path (containing JPG or PNG images)"
- echo " 2. Output dir path"
- echo " 3. Output CSV file path"
+ echo " -i Input dir path (containing JPG or PNG images)"
+ echo " -o Output dir path"
+ echo " -c Output CSV file path"
+ echo " -l Path to Keras LP detector model (default = $lp_model)"
+ echo " -h Print this help information"
  echo ""
  exit 1
-fi
-
-read -n1 -r -p "Press any key to continue..." key
-
+}
 
-# Download all networks
-bash get-networks.sh
+while getopts 'i:o:c:l:h' OPTION; do
+ case $OPTION in
+ i) input_dir=$OPTARG;;
+ o) output_dir=$OPTARG;;
+ c) csv_file=$OPTARG;;
+ l) lp_model=$OPTARG;;
+ h) usage;;
+ esac
+done
 
+if [ -z "$input_dir" ]; then echo "Input dir not set."; usage; exit 1; fi
+if [ -z "$output_dir" ]; then echo "Ouput dir not set."; usage; exit 1; fi
+if [ -z "$csv_file" ]; then echo "CSV file not set." ; usage; exit 1; fi
 
 # Check if input dir exists
-check_dir $1
+check_dir $input_dir
 retval=$?
 if [ $retval -eq 0 ]
 then
- echo "Input directory ($1) does not exist"
- exit 0
+ echo "Input directory ($input_dir) does not exist"
+ exit 1
 fi
 
 # Check if output dir exists, if not, create it
-check_dir $2
+check_dir $output_dir
 retval=$?
 if [ $retval -eq 0 ]
 then
- mkdir -p $2
+ mkdir -p $output_dir
 fi
 
+# End if any error occur
+set -e
+
 # Detect vehicles
-python vehicle-detection.py $1 $2
+python vehicle-detection.py $input_dir $output_dir
 
 # Detect license plates
-python license-plate-detection.py $2 data/lp-detector/wpod-net.h5
+python license-plate-detection.py $output_dir $lp_model
 
 # OCR
-python license-plate-ocr.py $2
+python license-plate-ocr.py $output_dir
 
 # Draw output and generate list
-python gen-outputs.py $1 $2 > $3
+python gen-outputs.py $input_dir $output_dir > $csv_file
 
 # Clean files and draw output
-rm $2/*_lp.png
-rm $2/*car.png
-rm $2/*_cars.txt
-rm $2/*_lp.txt
-rm $2/*_str.txt
+rm $output_dir/*_lp.png
+rm $output_dir/*car.png
+rm $output_dir/*_cars.txt
+rm $output_dir/*_lp.txt
+rm $output_dir/*_str.txt