def test_saved_model_fp16(self):
self.params['hparams'] = 'mixed_precision=true'
inspector = model_inspect.ModelInspector(**self.params)
inspector.run_model('saved_model')
self.assertTrue(
os.path.exists(os.path.join(self.savedmodel_dir, 'saved_model.pb')))
utils.set_precision_policy('float32')
def test_saved_model_infer_dynamic_batch(self):
# Build saved model with dynamic batch size.
self.params['batch_size'] = None
inspector = model_inspect.ModelInspector(**self.params)
inspector.run_model('saved_model')
outdir = os.path.join(self.tempdir, 'infer_imgout_dyn')
os.mkdir(outdir)
img_path = os.path.join(self.tempdir, 'img.jpg')
Image.fromarray(self.test_image).save(img_path)
test_image2 = np.random.randint(0, 244, (640, 320, 3)).astype(np.uint8)
img2_path = os.path.join(self.tempdir, 'img2.jpg')
Image.fromarray(test_image2).save(img2_path)
# serve images with batch size 1.
tf.reset_default_graph()
self.params['batch_size'] = 1
self.assertFalse(os.path.exists(os.path.join(outdir, '0.jpg')))
inspector.run_model('saved_model_infer',
input_image=img_path,
output_image_dir=outdir)
self.assertTrue(os.path.exists(os.path.join(outdir, '0.jpg')))
# serve images with batch size 2.
tf.reset_default_graph()
self.params['batch_size'] = 2
self.assertFalse(os.path.exists(os.path.join(outdir, '1.jpg')))
fname = img_path.replace('img.jpg', 'img*.jpg')
inspector.run_model('saved_model_infer',
input_image=fname,
output_image_dir=outdir)
self.assertTrue(os.path.exists(os.path.join(outdir, '1.jpg')))
def test_saved_model(self):
inspector = model_inspect.ModelInspector(**self.params)
self.assertFalse(
os.path.exists(os.path.join(self.savedmodel_dir, 'saved_model.pb')))
inspector.run_model('saved_model')
self.assertTrue(
os.path.exists(os.path.join(self.savedmodel_dir, 'saved_model.pb')))
self.assertTrue(
os.path.exists(
os.path.join(self.savedmodel_dir, 'efficientdet-d0_frozen.pb')))
def test_infer(self):
outdir = os.path.join(self.tempdir, 'infer_imgout')
os.mkdir(outdir)
inspector = model_inspect.ModelInspector(**self.params)
img_path = os.path.join(self.tempdir, 'img.jpg')
Image.fromarray(self.test_image).save(img_path)
self.assertFalse(os.path.exists(os.path.join(outdir, '0.jpg')))
inspector.run_model('infer', input_image=img_path, output_image_dir=outdir)
self.assertTrue(os.path.exists(os.path.join(outdir, '0.jpg')))
out = np.sum(np.array(Image.open(os.path.join(outdir, '0.jpg'))))
self.assertEqual(out // 10000000, 16)
def test_saved_model(self):
if tf.__version__ >= '2.3.0-dev20200521':
self.params['tflite_path'] = os.path.join(self.savedmodel_dir, 'x.tflite')
inspector = model_inspect.ModelInspector(**self.params)
self.assertFalse(
os.path.exists(os.path.join(self.savedmodel_dir, 'saved_model.pb')))
inspector.run_model('saved_model')
self.assertTrue(
os.path.exists(os.path.join(self.savedmodel_dir, 'saved_model.pb')))
self.assertTrue(
os.path.exists(
os.path.join(self.savedmodel_dir, 'efficientdet-d0_frozen.pb')))
if self.params.get('tflite_path', None):
self.assertTrue(
os.path.exists(os.path.join(self.savedmodel_dir, 'x.tflite')))
def test_saved_model_graph_infer(self):
inspector = model_inspect.ModelInspector(**self.params)
inspector.run_model('saved_model')
tf.reset_default_graph()
# Use the frozen graph to do inference.
inspector.saved_model_dir = os.path.join(self.params['saved_model_dir'],
'efficientdet-d0_frozen.pb')
outdir = os.path.join(self.tempdir, 'pb_infer_imgout')
os.mkdir(outdir)
img_path = os.path.join(self.tempdir, 'img.jpg')
Image.fromarray(self.test_image).save(img_path)
self.assertFalse(os.path.exists(os.path.join(outdir, '0.jpg')))
inspector.run_model(
'saved_model_infer', input_image=img_path, output_image_dir=outdir)
self.assertTrue(os.path.exists(os.path.join(outdir, '0.jpg')))
out = np.sum(np.array(Image.open(os.path.join(outdir, '0.jpg'))))
self.assertEqual(out // 10000000, 16)
def test_eval_ckpt(self):
inspector = model_inspect.ModelInspector(**self.params)
inspector.run_model('ckpt')
def test_bm(self):
inspector = model_inspect.ModelInspector(**self.params)
inspector.run_model('bm')
def test_freeze_model(self):
inspector = model_inspect.ModelInspector(**self.params)
inspector.run_model('freeze')
def test_dry_run(self):
inspector = model_inspect.ModelInspector(**self.params)
inspector.run_model('dry')
import random
import os
import tensorflow.compat.v1 as tf
import numpy as np
import itertools
import model_inspect
from waymo_open_dataset.utils import range_image_utils
from waymo_open_dataset.utils import transform_utils
from waymo_open_dataset.utils import frame_utils
from waymo_open_dataset import dataset_pb2
from waymo_open_dataset import label_pb2
from waymo_open_dataset.protos import metrics_pb2
inspector = model_inspect.ModelInspector('efficientdet-d0', 'logdir')
filepath = '/mnt/pool0-100/shape-it/validation_data/segment-14486517341017504003_3406_349_3426_349_with_camera_labels.tfrecord'
dataset = tf.data.TFRecordDataset(filepath, compression_type='')
numFrames = 0
config_dict = {}
config_dict['line_thickness'] = 3
config_dict['max_boxes_to_draw'] = 15
config_dict['min_score_thresh'] = 0.5
for data in dataset:
if numFrames >= 3:
break
numFrames = numFrames + 1
frame = dataset_pb2.Frame()
frame.ParseFromString(bytearray(data.numpy()))
numImages = 0
