Fix TF 2.x compatibility and Add code to get object labels

annotations/annotations_test.csv

@@ -0,0 +1,3 @@
+test1.jpg,0,0,10,10,object,3000,2000
+test2.jpg,0,0,10,10,object,427,367
+test3.jpg,0,0,10,10,object,363,481

object_detector_retinanet/keras_retinanet/backend/tensorflow_backend.py

@@ -53,7 +53,7 @@ def resize_images(images, size, method='bilinear', align_corners=False):
         'bicubic' : tensorflow.image.ResizeMethod.BICUBIC,
         'area'    : tensorflow.image.ResizeMethod.AREA,
     }
-    return tensorflow.image.resize_images(images, size, methods[method], align_corners)
+    return tensorflow.image.resize(images, size, methods[method], align_corners)
 
 
 def non_max_suppression(*args, **kwargs):

object_detector_retinanet/keras_retinanet/bin/predict.py

@@ -20,7 +20,7 @@
 import os
 import sys
 
-import keras
+import tensorflow.keras as keras
 import tensorflow as tf
 
 
@@ -34,9 +34,9 @@
 def get_session():
     """ Construct a modified tf session.
     """
-    config = tf.ConfigProto()
+    config = tf.compat.v1.ConfigProto()
     config.gpu_options.allow_growth = True
-    return tf.Session(config=config)
+    return tf.compat.v1.Session(config=config)
 
 
 def create_generator(args):
@@ -127,7 +127,7 @@ def main(args=None):
         os.environ["CUDA_VISIBLE_DEVICES"] = str(666)
     else:
         os.environ["CUDA_VISIBLE_DEVICES"] = gpu_num
-    keras.backend.tensorflow_backend.set_session(get_session())
+    tf.compat.v1.keras.backend.set_session(get_session())
 
     # make save path if it doesn't exist
     if args.save_path is not None and not os.path.exists(args.save_path):

object_detector_retinanet/keras_retinanet/initializers.py

@@ -34,6 +34,6 @@ def get_config(self):
 
     def __call__(self, shape, dtype=None):
         # set bias to -log((1 - p)/p) for foreground
-        result = np.ones(shape, dtype=dtype) * -math.log((1 - self.probability) / self.probability)
+        result = np.ones(shape, dtype=np.float32) * -math.log((1 - self.probability) / self.probability)
 
         return result

object_detector_retinanet/keras_retinanet/layers/filter_detections.py

@@ -14,12 +14,12 @@
 limitations under the License.
 """
 
-import keras
+import tensorflow as tf
+import tensorflow.keras as keras
 import numpy
 
 from .. import backend
 
-
 def filter_detections(boxes, classification, other=[], nms=True, score_threshold=0.05, max_detections=300, nms_threshold=0.5):
     """ Filter detections using the boxes and classification values.
 
@@ -60,9 +60,11 @@ def filter_detections(boxes, classification, other=[], nms=True, score_threshold
             indices = keras.backend.gather(indices, nms_indices)
 
         # add indices to list of all indices
-        labels  = c * keras.backend.ones((keras.backend.shape(indices)[0],), dtype='int64')
+        init_shape = (keras.backend.shape(indices)[0],)
+        labels  = c * tf.ones(init_shape, dtype='int64')
         indices = keras.backend.stack([indices[:, 0], labels], axis=1)
         all_indices.append(indices)
+
 
     # concatenate indices to single tensor
     indices = keras.backend.concatenate(all_indices, axis=0)

object_detector_retinanet/keras_retinanet/models/__init__.py

@@ -71,9 +71,9 @@ def load_model(filepath, backbone_name='resnet50', convert=False, nms=False):
         ImportError: if h5py is not available.
         ValueError: In case of an invalid savefile.
     """
-    import keras.models
+    import tensorflow.keras.models
 
-    model = keras.models.load_model(filepath, custom_objects=backbone(backbone_name).custom_objects)
+    model = tensorflow.keras.models.load_model(filepath, custom_objects=backbone(backbone_name).custom_objects)
     if convert:
         from object_detector_retinanet.keras_retinanet.models.retinanet import retinanet_bbox
         model = retinanet_bbox(model=model, nms=nms)

object_detector_retinanet/keras_retinanet/models/densenet.py

@@ -14,7 +14,7 @@
 limitations under the License.
 """
 
-import keras
+import tensorflow.keras as keras
 from keras.applications.densenet import DenseNet, get_file
 
 from . import retinanet

object_detector_retinanet/keras_retinanet/models/mobilenet.py

@@ -14,7 +14,7 @@
 limitations under the License.
 """
 
-import keras
+import tensorflow.keras as keras
 from keras.applications.mobilenet import MobileNet, BASE_WEIGHT_PATH, get_file, relu6, DepthwiseConv2D
 
 from . import retinanet

object_detector_retinanet/keras_retinanet/models/resnet.py

@@ -14,7 +14,7 @@
 limitations under the License.
 """
 
-import keras
+import tensorflow.keras as keras
 import keras_resnet
 import keras_resnet.models
 from keras.utils import get_file

object_detector_retinanet/keras_retinanet/models/retinanet.py

@@ -14,7 +14,7 @@
 limitations under the License.
 """
 
-import keras
+import tensorflow.keras as keras
 from .. import initializers
 from .. import layers
 

object_detector_retinanet/keras_retinanet/preprocessing/generator.py

@@ -19,7 +19,7 @@
 import threading
 import warnings
 
-import keras
+import tensorflow.keras as keras
 
 from ..utils.anchors import anchor_targets_bbox, bbox_transform
 from ..utils.image import (

object_detector_retinanet/keras_retinanet/utils/Boxes.py

@@ -145,7 +145,7 @@ def perform_nms_on_image_dataframe(image_data, overlap_threshold=0.5):
     number_of_images = len(image_data['image_name'].unique())
     if number_of_images > 1:
         print('nms received data including more than 1 image - cannot perform nms!')
-    image_boxes = image_data.as_matrix(BOX_CONSTANTS)
+    image_boxes = image_data.to_numpy()
     image_scores = numpy.array(image_data['confidence'])
 
     nms_boxes, nms_scores, deleted_indices = non_maximal_suppression(image_boxes, image_scores,

object_detector_retinanet/keras_retinanet/utils/CollapsingMoG.py

@@ -15,11 +15,13 @@ def __init__(self, sec):
         self.sec = sec
 
     def __enter__(self):
-        signal.signal(signal.SIGALRM, self.raise_timeout)
-        signal.setitimer(signal.ITIMER_REAL, self.sec)
+        # signal.signal(signal.SIGALRM, self.raise_timeout)
+        # signal.setitimer(signal.ITIMER_REAL, self.sec)
+        pass
 
     def __exit__(self, *args):
-        signal.alarm(0)  # disable alarm
+        # signal.alarm(0)  # disable alarm
+        pass
 
     def raise_timeout(self, *args):
         raise Timeout.Timeout()

object_detector_retinanet/keras_retinanet/utils/EmMerger.py

@@ -64,6 +64,8 @@ def filter_duplicate_candidates(self, data, image):
         cv2.normalize(heat_map, heat_map, 0, 255, cv2.NORM_MINMAX)
         heat_map = cv2.convertScaleAbs(heat_map)
         h2, heat_map = cv2.threshold(heat_map, 4, 255, cv2.THRESH_TOZERO)
+        # cv2.imshow("Heat map", heat_map)
+        # cv2.waitKey()
         contours = cv2.findContours(numpy.ndarray.copy(heat_map), cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
 
         candidates = self.find_new_candidates(contours, heat_map, data, original_detection_centers, image)
@@ -75,13 +77,12 @@ def filter_duplicate_candidates(self, data, image):
         filtered_data = pandas.DataFrame(columns=data.columns)
         for i, candidate in candidates.items():
             label = candidate['original_detection_ids']
-            original_detections = data.ix[label]
-            original_detections[
-                'avg_score'] = 0.5 * original_detections.confidence + 0.5 * original_detections.hard_score
+            original_detections = data.iloc[label]
+            original_detections.insert(0, 'avg_score', 0.5 * original_detections.confidence + 0.5 * original_detections.hard_score, True)
             best_detection_id = original_detections.avg_score.argmax()
             # best_detection_id = original_detections.confidence.argmax()
             # best_detection_id = original_detections.hard_score.argmax()
-            best_detection = original_detections.ix[best_detection_id].copy()
+            best_detection = original_detections.iloc[best_detection_id].copy()
 
             # The following code creates the median bboxes
             # original_detections = original_detections[original_detections.confidence > 0.5]
@@ -110,7 +111,7 @@ def filter_duplicate_candidates(self, data, image):
 
     def find_new_candidates(self, contours, heat_map, data, original_detection_centers, image):
         candidates = []
-        for contour_i, contour in enumerate(contours[1]):
+        for contour_i, contour in enumerate(contours[0]):
             contour_bounding_rect = cv2.boundingRect(contour)
 
             contour_bbox = extract_boxes_from_edge_boxes(numpy.array(contour_bounding_rect))[0]
@@ -200,8 +201,9 @@ def remove_redundant(self, contour_bbox, cov, k, mu, image, sub_heat_map):
             ellipse_mask = cv2.fillPoly(local_m, [poly], (1, 1, 1))
             contours = cv2.findContours(ellipse_mask.copy(), cv2.RETR_EXTERNAL,
                                         cv2.CHAIN_APPROX_SIMPLE)
-            cnts.append(contours[1][0])
+            cnts.append(contours[0][0])
         center_points = mu.copy()
+
         distances = scipy.spatial.distance.cdist(center_points, center_points)
         scaled_distances = numpy.ndarray(shape=[k, k], dtype=numpy.float64)
         for i in range(0, k):
@@ -368,6 +370,7 @@ def merge_detections(image_name, results):
     result_df['y2'] = results[:, 3].astype(int)
     result_df['confidence'] = results[:, 4]
     result_df['hard_score'] = results[:, 5]
+    result_df['classification'] = results[:, 6]
     result_df['uuid'] = 'object_label'
     result_df['label_type'] = 'object_label'
 #    result_df['project'] = project

object_detector_retinanet/keras_retinanet/utils/image.py

@@ -15,7 +15,7 @@
 """
 
 from __future__ import division
-import keras
+import tensorflow.keras as keras
 import numpy as np
 import cv2
 from PIL import Image

object_detector_retinanet/keras_retinanet/utils/keras_version.py

@@ -16,7 +16,7 @@
 
 from __future__ import print_function
 
-import keras
+import tensorflow.keras as keras
 import sys
 
 minimum_keras_version = 2, 1, 3

object_detector_retinanet/keras_retinanet/utils/predict_iou.py

@@ -39,11 +39,11 @@ def predict(
         hard_score_rate=1.):
     all_detections = [[None for i in range(generator.num_classes())] for j in range(generator.size())]
     csv_data_lst = []
-    csv_data_lst.append(['image_id', 'x1', 'y1', 'x2', 'y2', 'confidence', 'hard_score'])
+    csv_data_lst.append(['image_id', 'x1', 'y1', 'x2', 'y2', 'confidence', 'hard_score', 'classification'])
     result_dir = os.path.join(root_dir(), 'results')
     create_folder(result_dir)
     timestamp = datetime.datetime.utcnow()
-    res_file = result_dir + '/detections_output_iou_{}_{}.csv'.format(hard_score_rate, timestamp)
+    res_file = result_dir + '/detections_output_iou_{}.csv'.format(hard_score_rate)
     for i in range(generator.size()):
         image_name = os.path.join(generator.image_path(i).split(os.path.sep)[-2],
                                   generator.image_path(i).split(os.path.sep)[-1])
@@ -89,7 +89,7 @@ def predict(
             filtered_scores.append(detection['confidence'])
             filtered_labels.append('{0:.2f}'.format(detection['hard_score']))
             row = [image_name, detection['x1'], detection['y1'], detection['x2'], detection['y2'],
-                   detection['confidence'], detection['hard_score']]
+                   detection['confidence'], detection['hard_score'], generator.labels[int(detection['classification'])]]
             csv_data_lst.append(row)
 
         if save_path is not None:

object_detector_retinanet/utils.py

@@ -40,7 +40,7 @@ def root_dir():
     if platform.system() == 'Linux':
         return os.path.join(os.getenv('HOME'), 'Documents', 'SKU110K')
     elif platform.system() == 'Windows':
-        return os.path.abspath('C:/Users/{}/Documents/SKU110K/'.format(os.getenv('username')))
+        return os.path.abspath('C:/Users/{}/Desktop/SKU110K_CVPR19/'.format(os.getenv('username')))
 
 
 def image_path():

requirements.txt

@@ -1,7 +1,8 @@
-Keras==2.2.5
-keras-retinanet==0.5.1
-tensorflow-gpu==1.15.4
-numpy==1.19.2
-opencv-python==3.1.0.5
-tqdm==4.50.2
-pandas==0.23.4
+keras==2.4.3
+keras_resnet
+tensorflow==2.4.0
+numpy==1.19.5
+opencv-python==4.5.1
+opencv-contrib-python==4.5.1
+tqdm
+pandas