I am new to TensorFlow and I've been trying to figure out how the object detection API works.
I've been following this tutorial. I've made a few changes to pipeline.config that deals with learning rate,
post_processing > batch_non_max_suppression >
score_threshold: 0.165
iou_threshold: 0.49000000238418579
max_detections_per_class: 5
max_total_detections: 10
use_static_shapes: false
And here are my graphs>
tensorboard_graphs
After around 700+ steps when my total loss converged to around 1. So, I stopped the training and then exported the saved model.
For training, exporting and inference I've used the code provided here.
model_main_tf2.py
exporter_main_v2.py
inference_from_saved_model_tf2_colab.ipynb
My model is trained with 1425 images of pistols obtained from this website. It is trained for one class(PS I was thinking that for object detection, as there is classification and localization having one class is alright because parts outside the bounding box is negative. Let me know if I'm wrong here.)
print(detections['detection_anchor_indices'])
tf.Tensor([[0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]], shape=(1, 10), dtype=float32)
print(detections['detection_boxes'])
tf.Tensor(
[[[0. 0. 0. 0.]
[0. 0. 0. 0.]
[0. 0. 0. 0.]
[0. 0. 0. 0.]
[0. 0. 0. 0.]
[0. 0. 0. 0.]
[0. 0. 0. 0.]
[0. 0. 0. 0.]
[0. 0. 0. 0.]
[0. 0. 0. 0.]]], shape=(1, 10, 4), dtype=float32)
print(detections['detection_classes'])
tf.Tensor([[1. 1. 1. 1. 1. 1. 1. 1. 1. 1.]], shape=(1, 10), dtype=float32)
print(detections['detection_multiclass_scores'])
tf.Tensor(
[[[0. 0.]
[0. 0.]
[0. 0.]
[0. 0.]
[0. 0.]
[0. 0.]
[0. 0.]
[0. 0.]
[0. 0.]
[0. 0.]]], shape=(1, 10, 2), dtype=float32)
print(detections['detection_scores'])
tf.Tensor([[0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]], shape=(1, 10), dtype=float32)
print(detections['num_detections'])
tf.Tensor([0.], shape=(1,), dtype=float32)
I have checked if the dataset was labeled properly with the help of labelImg. I dont know if I should train it more or is it a fault with my dataset. I am afraid if it would overfit. My wildest guess right now is if I have done something wrong at creating the TFRecord.
def create_tf_example(group, path):
with tf.gfile.GFile(os.path.join(path, '{}'.format(group.filename)), 'rb') as fid:
encoded_jpg = fid.read()
encoded_jpg_io = io.BytesIO(encoded_jpg)
image = Image.open(encoded_jpg_io)
width, height = image.size
filename = group.filename.encode('utf8')
image_format = b'jpg'
xmins = []
xmaxs = []
ymins = []
ymaxs = []
classes_text = []
classes = []
for index, row in group.object.iterrows():
xmins.append(row['xmin'] / width)
xmaxs.append(row['xmax'] / width)
ymins.append(row['ymin'] / height)
ymaxs.append(row['ymax'] / height)
classes_text.append(row['class'].encode('utf8'))
classes.append(class_text_to_int(row['class']))
tf_example = tf.train.Example(features=tf.train.Features(feature={
'image/height': dataset_util.int64_feature(height),#tf.io.FixedLenFeature((), tf.int64),
'image/width': dataset_util.int64_feature(width),
'image/filename': dataset_util.bytes_feature(filename),
'image/source_id': dataset_util.bytes_feature(filename),
'image/encoded': dataset_util.bytes_feature(encoded_jpg),
'image/format': dataset_util.bytes_feature(image_format),
'image/object/bbox/xmin': dataset_util.float_list_feature(xmins),
'image/object/bbox/xmax': dataset_util.float_list_feature(xmaxs),
'image/object/bbox/ymin': dataset_util.float_list_feature(ymins),
'image/object/bbox/ymax': dataset_util.float_list_feature(ymaxs),
'image/object/class/text': dataset_util.bytes_list_feature(classes_text),
'image/object/class/label': dataset_util.int64_list_feature(classes),
}))
return tf_example
Any help would be appreciated. Btw I am using mobilenet v2 ssd model as my pre-trained model.
And just to assure is this the best way to make a handy object detection model?
