def vidvrd_generator(self):
feature_type = 'train'
# for each_ins in vid_data.gen_vrd_instance(feature_type):
# yield {
# "image/encoded": np.array2string(each_ins.get_my_feature(feature_type).ravel()),
# "image/format": ["jpeg"],
# "image/class/label": [vid_relation.load_relation('first')[each_ins.predicate.split('_')[0]]],
# "image/height": [each_ins.height],
# "image/width": [each_ins.width]
# }
ins_list = vid_data.gen_vrd_instance(feature_type)
for i in range(3):
yield {
"image/encoded":
np.array2string(
ins_list[i].get_my_feature(feature_type).ravel()),
"image/format": ["jpeg"],
"image/class/label": [
vid_relation.load_relation('first')[
ins_list[i].predicate.split('_')[0]]
],
"image/height": [ins_list[i].height],
"image/width": [ins_list[i].width]
}
def feature_encoders(self, data_dir):
del data_dir
feature_type = 'train'
for each_ins in vid_data.gen_vrd_instance(feature_type):
yield {
"inputs": each_ins.get_my_feature(feature_type),
"targets": vid_relation.load_relation('first')[each_ins.predicate.split('_')[0]]
}
def vidvrd_generator(self):
feature_type = 'train'
for each_ins in vid_data.gen_vrd_instance(feature_type):
yield {
"image/encoded": np.array2string(each_ins.get_my_feature(feature_type).ravel()),
"image/format": ["png"],
"image/class/label": [vid_relation.load_relation('first')[each_ins.predicate.split('_')[0]]],
"image/height": [each_ins.height],
"image/width": [each_ins.width]
}
class FrameClass(image_utils.Image2ClassProblem):
# feature_path = '../data/VidVRD-features/vid_features'
classes_num = len(vid_relation.load_relation('first'))
@property
def num_generate_tasks(self):
return self.classes_num
def prepare_to_generate(self, data_dir, tmp_dir):
pass
def generator(self, data_dir, tmp_dir, is_training):
return self.vidvrd_generator()
@property
def num_channels(self):
return 3
@property
def is_small(self):
return False
@property
def num_classes(self):
return self.classes_num
@property
def class_labels(self):
return vid_relation.load_relation(self.relation_path)
@property
def train_shards(self):
return self.classes_num
def feature_encoders(self, data_dir):
del data_dir
feature_type = 'train'
for each_ins in vid_data.gen_vrd_instance(feature_type):
yield {
"inputs": each_ins.get_my_feature(feature_type),
"targets": vid_relation.load_relation('first')[each_ins.predicate.split('_')[0]]
}
def vidvrd_generator(self):
feature_type = 'train'
for each_ins in vid_data.gen_vrd_instance(feature_type):
yield {
"image/encoded": np.array2string(each_ins.get_my_feature(feature_type).ravel()),
"image/format": ["png"],
"image/class/label": [vid_relation.load_relation('first')[each_ins.predicate.split('_')[0]]],
"image/height": [each_ins.height],
"image/width": [each_ins.width]
}
def class_labels(self):
return vid_relation.load_relation(self.relation_path)
"targets": vid_relation.load_relation('first')[each_ins.predicate.split('_')[0]]
}
def vidvrd_generator(self):
feature_type = 'train'
for each_ins in vid_data.gen_vrd_instance(feature_type):
yield {
"image/encoded": np.array2string(each_ins.get_my_feature(feature_type).ravel()),
"image/format": ["png"],
"image/class/label": [vid_relation.load_relation('first')[each_ins.predicate.split('_')[0]]],
"image/height": [each_ins.height],
"image/width": [each_ins.width]
}
# ins_list = vid_data.gen_vrd_instance(feature_type)
# for i in range(3):
# yield {
# "image/encoded": np.array2string(ins_list[i].get_my_feature(feature_type).ravel()),
# "image/format": ["png"],
# "image/class/label": [vid_relation.load_relation('first')[ins_list[i].predicate.split('_')[0]]],
# "image/height": [ins_list[i].height],
# "image/width": [ins_list[i].width]
# }
if __name__ == '__main__':
feature_type = 'train'
each_ins = vid_data.gen_vrd_instance(feature_type)[0]
print(each_ins.get_my_feature(feature_type))
print(vid_relation.load_relation('first')[each_ins.predicate.split('_')[0]])