2martens/masterthesis
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

Updated predict functionality to use Keras

Browse Source 
Signed-off-by: Jim Martens <github@2martens.de>
This commit is contained in:
Jim Martens
2019-06-13 16:56:12 +02:00
parent d466e8e376
commit b2a1db6485
1 changed files with 20 additions and 16 deletions
Download Patch File Download Diff File Expand all files Collapse all files

34
src/twomartens/masterthesis/cli.py
View File

@ -249,7 +249,6 @@ def _ssd_val(args: argparse.Namespace) -> None:
import os import os
import tensorflow as tf import tensorflow as tf
from tensorflow.python.ops import summary_ops_v2
from twomartens.masterthesis import data from twomartens.masterthesis import data
from twomartens.masterthesis import ssd from twomartens.masterthesis import ssd
@ -260,12 +259,12 @@ def _ssd_val(args: argparse.Namespace) -> None:
tf.enable_eager_execution(config=config) tf.enable_eager_execution(config=config)
batch_size = 16 batch_size = 16
image_size = 300 image_size = (300, 300)
forward_passes_per_image = 10 forward_passes_per_image = 10
use_dropout = False if args.network == "ssd" else True use_dropout = False if args.network == "ssd" else True
weights_file = f"{args.weights_path}/VGG_coco_SSD_300x300_iter_400000.h5"
checkpoint_path = f"{args.weights_path}/train/{args.network}/{args.train_iteration}" checkpoint_path = f"{args.weights_path}/train/{args.network}/{args.train_iteration}"
model_file = f"{checkpoint_path}/ssd300.h5"
output_path = f"{args.output_path}/val/{args.network}/{args.iteration}/" output_path = f"{args.output_path}/val/{args.network}/{args.iteration}/"
os.makedirs(output_path, exist_ok=True) os.makedirs(output_path, exist_ok=True)
@ -275,22 +274,27 @@ def _ssd_val(args: argparse.Namespace) -> None:
with open(f"{args.ground_truth_path}/instances.bin", "rb") as file: with open(f"{args.ground_truth_path}/instances.bin", "rb") as file:
instances = pickle.load(file) instances = pickle.load(file)
scenenet_data, nr_digits, length_dataset = \ # model
ssd_model = tf.keras.models.load_model(model_file)
test_generator, length_dataset = \
data.load_scenenet_data(file_names_photos, instances, args.coco_path, data.load_scenenet_data(file_names_photos, instances, args.coco_path,
predictor_sizes=ssd_model.predictor_sizes,
batch_size=batch_size, batch_size=batch_size,
resized_shape=(image_size, image_size)) resized_shape=image_size,
training=False, evaluation=True)
del file_names_photos, instances del file_names_photos, instances
use_summary_writer = summary_ops_v2.create_file_writer( nr_digits = math.ceil(math.log10(math.ceil(length_dataset / batch_size)))
f"{args.summary_path}/val/{args.network}/{args.iteration}" steps_per_epoch = int(math.ceil(length_dataset / batch_size))
) ssd.predict_keras(test_generator,
if args.debug: steps_per_epoch,
with use_summary_writer.as_default(): ssd_model,
ssd.predict(scenenet_data, use_dropout, output_path, weights_file, checkpoint_path, use_dropout,
nr_digits=nr_digits, forward_passes_per_image=forward_passes_per_image) forward_passes_per_image,
else: image_size,
ssd.predict(scenenet_data, use_dropout, output_path, weights_file, checkpoint_path, output_path,
nr_digits=nr_digits, forward_passes_per_image=forward_passes_per_image) nr_digits)
def _auto_encoder_val(args: argparse.Namespace) -> None: def _auto_encoder_val(args: argparse.Namespace) -> None: