yeephycho / widerface-to-tfrecord

parse WIDER FACE dataset to tensorflow tfrecord format for object detection api

widerface-to-tfrecord

parse WIDER FACE dataset to tensorflow tfrecord format for object detection api.

Dependencies

numpy

opencv 2.7

tensorflow > 1.0.1

Usage

Enter project root

git clone https://github.com/yeephycho/widerface-to-tfrecord.git
cd ~/widerface-to-tfrecord

Create simbolic link to WIDER FACE dataset

Unzip .zip file to folder "~/WIDER/WIDER_train" and make sure "wider_face_train_annot.txt" is under folder "~/WIDER".

ln -s ~/WIDER ./widerface-to-tfrecord/WIDER

Run the script

python widerface_To_TFRecord.py

License

Apache 2.0

Auther

Yeephycho

widerface-to-tfrecord's Issues

what is ln in windows 8.1 ?

Ln command does not exist in windows 8.1 . what is the alternative command for it?

New Python3 port!

Hi!

Based on your work, I've released a Python3 port with some minor improvements!

Check it out here if you are interested.

I leave this issue here in case someone drops by here and needs the Python3 version.

Thanks for this work!

Some modifications needed

There was some errors.

1. location of annotation

f = open("WIDER/wider_face_train_annot.txt")

I noticed in the Face annotations came in different file.

This solved it:

f = open("WIDER/wider_face_split/wider_face_train_bbx_gt.txt")

2. Encoding issues

Some lines needed UTF-8 encoding in tf.train.Example() line

Original:

tf_example = tf.train.Example(features=tf.train.Features(feature={
'image/height': dataset_util.int64_feature(int(height)),
'image/width': dataset_util.int64_feature(int(width)),
'image/filename': dataset_util.bytes_feature(filename),
'image/source_id': dataset_util.bytes_feature(filename),
'image/key/sha256': dataset_util.bytes_feature(key.encode('utf8')),
'image/encoded': dataset_util.bytes_feature(encoded_image_data),
'image/format': dataset_util.bytes_feature('jpeg'.encode('utf8')),
'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),
'image/object/difficult': dataset_util.int64_list_feature(int(0)),
'image/object/truncated': dataset_util.int64_list_feature(truncated),
'image/object/view': dataset_util.bytes_list_feature(poses),
}))

Updated code

tf_example = tf.train.Example(features=tf.train.Features(feature={
'image/height': dataset_util.int64_feature(int(height)),
'image/width': dataset_util.int64_feature(int(width)),
'image/filename': dataset_util.bytes_feature(filename.encode('utf8')),
'image/source_id': dataset_util.bytes_feature(filename.encode('utf8')),
'image/key/sha256': dataset_util.bytes_feature(key.encode('utf8')),
'image/encoded': dataset_util.bytes_feature(encoded_image_data),
'image/format': dataset_util.bytes_feature('jpeg'.encode('utf8')),
'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([x.encode('utf8') for x in classes_text]),
'image/object/class/label': dataset_util.int64_list_feature(classes),
'image/object/difficult': dataset_util.int64_list_feature(int(0)),
'image/object/truncated': dataset_util.int64_list_feature(truncated),
'image/object/view': dataset_util.bytes_list_feature(poses),
}))

Logic of this condition

I don't understand the logic of these lines. Why are you hardcoding these? Why should the minimum bbox be 25 and 30?

widerface-to-tfrecord/widerface_To_TFRecord.py

Lines 55 to 56 in ae945f4

	if(float(annot[2]) > 25.0):
	if(float(annot[3]) > 30.0):