oxford dataset ready to train

egohands_setup.py

@@ -241,6 +241,7 @@ def create_label_map():
         f = open(label_map,"w")
         f.write("item {\n  id: 1\n  name: 'hand'\n}")
         f.close()
+    print("> created ",label_map)
 
 def final_finish():
     cwd = os.getcwd()

howto_tf

@@ -11,14 +11,14 @@ export PYTHONPATH=$PYTHONPATH:`pwd`:`pwd`/slim
 # From the tensorflow/models/research directory
 python object_detection/train.py \
     --logtostderr \
-    --pipeline_config_path=/home/gustav/workspace/deeptraining_hands/model/ssd_mobilehandsnet.config \
+    --pipeline_config_path=/home/gustav/workspace/deeptraining_hands/model/ssd_mobilehandsnet_train.config \
     --train_dir=/home/gustav/workspace/deeptraining_hands/model/train
 
 ## EVALUATION TENSORFLOW
 # From the tensorflow/models/research directory
 python object_detection/eval.py \
     --logtostderr \
-    --pipeline_config_path=/home/gustav/workspace/deeptraining_hands/model/ssd_mobilehandsnet.config \
+    --pipeline_config_path=/home/gustav/workspace/deeptraining_hands/model/ssd_mobilehandsnet_train.config \
     --checkpoint_dir=/home/gustav/workspace/deeptraining_hands/model/train \
     --eval_dir=/home/gustav/workspace/deeptraining_hands/model/eval
 
@@ -29,9 +29,9 @@ tensorboard --logdir=/home/gustav/workspace/deeptraining_hands/model
 # From the tensorflow/models/research directory
 python object_detection/export_inference_graph.py \
 --input_type=image_tensor \
---pipeline_config_path=/home/gustav/workspace/deeptraining_hands/model/faster_rcnn_lowproposalhandsnet.config \
---trained_checkpoint_prefix /home/gustav/workspace/deeptraining_hands/model/train/model.ckpt-200000 \
---output_directory=/home/gustav/workspace/deeptraining_hands/model/frozen_model
+--pipeline_config_path=/home/gustav/workspace/deeptraining_hands/model/ssd_mobilehandsnet_train.config \
+--trained_checkpoint_prefix /home/gustav/workspace/deeptraining_hands/model/train/model.ckpt-149406 \
+--output_directory=/home/gustav/workspace/deeptraining_hands/model/train/frozen_model_149
 
 
 #### Options for the model.config file #### 

mat_to_xml.py

@@ -86,7 +86,8 @@ def read_mat_file(filepath, filename,IMG_FILES_PATH, XML_FILES_PATH):
 
     mat_data = sio.loadmat(filepath)
     hand_pos = [] # To store the hand positions in an image
-
+    old_pts = []
+    curr_pts = []
     #For all hands in the image align the bounding box to an axis
     for i in range(len(mat_data['boxes'][0])):
         xmin = float('inf')
@@ -95,7 +96,8 @@ def read_mat_file(filepath, filename,IMG_FILES_PATH, XML_FILES_PATH):
         ymax = -1
 
         for j in range(4):
-            x, y = mat_data['boxes'][0][i][0][0][j][0]
+            y, x = mat_data['boxes'][0][i][0][0][j][0]
+            curr_pts.append(make_int([x,y]))
             if xmin > x:
                 xmin = x
             if ymin > y:
@@ -104,8 +106,12 @@ def read_mat_file(filepath, filename,IMG_FILES_PATH, XML_FILES_PATH):
                 xmax = x
             if ymax < y:
                 ymax = y
-
+        old_pts.append(curr_pts)
         hand_pos.insert(0, [xmin, ymin, xmax, ymax])
+
+    img_filename = filename.split('.')[0] + '.jpg'
+    image = IMG_FILES_PATH + '/' + img_filename
+    visualize(image,old_pts,hand_pos,filename,1)
 
     # Create the XML file
     create_xml_file(hand_pos, filename,IMG_FILES_PATH, XML_FILES_PATH)
@@ -152,13 +158,32 @@ def create_xml_file(hand_pos, filename,IMG_FILES_PATH, XML_FILES_PATH):
     rough_xml = et.tostring(root, 'utf-8')
     rough_xml = minidom.parseString(rough_xml)
     pretty_xml = rough_xml.toprettyxml()
-    #print(pretty_xml)
 
     # Save the XML file
     xml_path = join(XML_FILES_PATH, xml_filename)
     with open(xml_path, 'w') as xml_file:
         xml_file.write(pretty_xml)
-
+
+def visualize(image,pts,boxes,filename,time):
+    font = cv2.FONT_HERSHEY_SIMPLEX
+    img = cv2.imread(image)
+
+    for pt in pts:
+        pt = np.array(pt)
+        cv2.polylines(img,[pt], True, (0, 0, 255), 2)
+    for box in boxes:
+        xmin, ymin, xmax, ymax = make_int(box)
+        cv2.rectangle(img, (xmin, ymin), (xmax, ymax), (0, 255, 0), 2)
+
+    cv2.putText(img, filename, (0,20), font, 0.75, (77, 255, 9), 2)
+    cv2.imshow('label_check', img)
+    cv2.waitKey(time)
+
+def make_int(box):
+    newbox = []
+    for pt in box:
+        newbox.append(int(pt))
+    return newbox
 
 
 def main():
@@ -175,12 +200,14 @@ def main():
         # List all files in the MAT_FILES_PATH and ignore hidden files (.DS_STORE for Macs)
         mat_files = [[join(MAT_FILES_PATH, x), x] for x in os.listdir(MAT_FILES_PATH) if isfile(join(MAT_FILES_PATH, x)) and x[0] is not '.']
         mat_files.sort()
+
         # Iterate through all files and convert them to XML
         for mat_file in mat_files:
             #print(mat_file)
             read_mat_file(mat_file[0], mat_file[1],IMG_FILES_PATH, XML_FILES_PATH)
             #break
         print ("successfully converted {}-labels from mat to pascal-xml").format(directory)
+    cv2.destroyAllWindows()
 
 if __name__ == '__main__':
     main()

model/ssd_mobilehandsnet.config → model/ssd_mobilehandsnet_train.config

@@ -181,11 +181,12 @@ train_input_reader: {
 }
 
 eval_config: {
-  num_examples: 480
+  num_examples: 1
+  eval_interval_secs: 600
   # Note: The below line limits the evaluation process to 10 evaluations.
   # Remove the below line to evaluate indefinitely.
   # max_evals: 10
-  visualization_export_dir: "/home/gustav/workspace/deeptraining_hands/model/eval/visualization"
+  # visualization_export_dir: "/home/gustav/workspace/deeptraining_hands/model/eval/visualization"
 }
 
 eval_input_reader: {

oxfordhands_setup.py

@@ -57,6 +57,7 @@ def create_label_map():
         f = open(label_map,"w")
         f.write("item {\n  id: 1\n  name: 'hand'\n}")
         f.close()
+    print("> created ",label_map)
 
 def cleanup_structure(data_path, dataset_path, tarfile_path):
     check = []