Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Efficientdet #230

Merged
merged 5 commits into from
Oct 27, 2022
Merged

Efficientdet #230

Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
70a0ace
Reference Efficientdet architectures are now saved as .hdf5
Manojkumarmuru Oct 25, 2022
97027b9
Dependence on .npy files removed for Pytest of anchor boxes
Manojkumarmuru Oct 25, 2022
9ad93b4
Testing of efficientdet architectures made independent of the .hdf5/j…
Manojkumarmuru Oct 26, 2022
a84e367
Test case to check the functionality of build_prior_boxes added
Manojkumarmuru Oct 26, 2022
8607225
Exception message modified.
Manojkumarmuru Oct 26, 2022
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
184 changes: 89 additions & 95 deletions examples/efficientdet/efficientdet_test.py
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,34 +1,24 @@
import numpy as np
# from examples.efficientdet.efficientdet_blocks import FeatureNode
import pytest
import tensorflow as tf
from tensorflow.keras import backend as K
from keras.utils.layer_utils import count_params
from tensorflow.keras.layers import Input
from tensorflow.keras.models import model_from_json

from anchors import build_prior_boxes
from efficientdet import (EFFICIENTDETD0, EFFICIENTDETD1, EFFICIENTDETD2,
EFFICIENTDETD3, EFFICIENTDETD4, EFFICIENTDETD5,
EFFICIENTDETD6, EFFICIENTDETD7)
from efficientdet_blocks import FuseFeature
from efficientnet_model import EfficientNet, conv_normal_initializer
from efficientnet_model import EfficientNet


@pytest.fixture
def models_base_path():
return ("/home/manummk95/Desktop/efficientdet_working/"
"required/test_files/efficientdet_architectures/")
def model_input_name():
return 'image'


@pytest.fixture
def model_input_output_base_path():
return ("/home/manummk95/Desktop/efficientdet_working/"
"required/test_files/test_model_outputs/")


@pytest.fixture
def model_anchor_boxes_path():
return ("/home/manummk95/Desktop/efficientdet_working/"
"required/test_files/efficientdet_anchors/")
def model_output_name():
return 'boxes'


def get_test_images(image_size, batch_size=1):
Expand Down Expand Up @@ -121,85 +111,89 @@ def test_efficientnet_features(input_shape, backbone, feature_shape,
del branch_tensors


@pytest.mark.parametrize('implemented_model, model_id',
[
(EFFICIENTDETD0, 0),
(EFFICIENTDETD1, 1),
(EFFICIENTDETD2, 2),
(EFFICIENTDETD3, 3),
(EFFICIENTDETD4, 4),
(EFFICIENTDETD5, 5),
(EFFICIENTDETD6, 6),
(EFFICIENTDETD7, 7),
])
def test_efficientdet_architecture(models_base_path,
implemented_model,
model_id):
custom_objects = {"conv_normal_initializer": conv_normal_initializer,
"FuseFeature": FuseFeature}
K.clear_session()
reference_model_path = (models_base_path + 'EFFICIENTDETD' +
str(model_id) + '.json')
reference_model_file = open(reference_model_path, 'r')
loaded_model_json = reference_model_file.read()
reference_model_file.close()
reference_model = model_from_json(loaded_model_json,
custom_objects=custom_objects)
K.clear_session()
assert (implemented_model().get_config() ==
reference_model.get_config()), ('EFFICIENTDETD' + str(model_id)
+ " architecture mismatch")
del implemented_model, reference_model


@pytest.mark.parametrize('model, model_id',
@pytest.mark.parametrize(('model, model_name, trainable_parameters,'
'non_trainable_parameters, input_shape,'
'output_shape'),
[
(EFFICIENTDETD0, 0),
(EFFICIENTDETD1, 1),
(EFFICIENTDETD2, 2),
(EFFICIENTDETD3, 3),
(EFFICIENTDETD4, 4),
(EFFICIENTDETD5, 5),
(EFFICIENTDETD6, 6),
(EFFICIENTDETD7, 7),
(EFFICIENTDETD0, 'efficientdet-d0', 3880067,
47136, (512, 512, 3), (49104, 94)),
(EFFICIENTDETD1, 'efficientdet-d1', 6625898,
71456, (640, 640, 3), (76725, 94)),
(EFFICIENTDETD2, 'efficientdet-d2', 8097039,
81776, (768, 768, 3), (110484, 94)),
(EFFICIENTDETD3, 'efficientdet-d3', 12032296,
114304, (896, 896, 3), (150381, 94)),
(EFFICIENTDETD4, 'efficientdet-d4', 20723675,
167312, (1024, 1024, 3), (196416, 94)),
(EFFICIENTDETD5, 'efficientdet-d5', 33653315,
227392, (1280, 1280, 3), (306900, 94)),
(EFFICIENTDETD6, 'efficientdet-d6', 51871934,
311984, (1280, 1280, 3), (306900, 94)),
(EFFICIENTDETD7, 'efficientdet-d7', 51871934,
311984, (1536, 1536, 3), (441936, 94)),
])
def test_efficientdet_anchor_boxes(model_anchor_boxes_path, model, model_id):
reference_anchor_file = (model_anchor_boxes_path + 'EFFICIENTDETD' +
str(model_id) + '_anchors.npy')
with open(reference_anchor_file, 'rb') as f:
reference_anchor = np.load(f)
model_anchor = model().prior_boxes
assert np.array_equal(model_anchor, reference_anchor)
del model


@pytest.mark.skip(reason="Training of model needs to carried out from D0-D7")
@pytest.mark.parametrize('model, model_idx, preprocessed_inputs',
def test_efficientdet_architecture(model, model_name, model_input_name,
model_output_name, trainable_parameters,
non_trainable_parameters, input_shape,
output_shape):
implemented_model = model()
trainable_count = count_params(
implemented_model.trainable_weights)
non_trainable_count = count_params(
implemented_model.non_trainable_weights)
assert implemented_model.name == model_name, "Model name incorrect"
assert implemented_model.input_names[0] == model_input_name, (
"Input name incorrect")
assert implemented_model.output_names[0] == model_output_name, (
"Output name incorrect")
assert trainable_count == trainable_parameters, (
"Incorrect trainable parameters count")
assert non_trainable_count == non_trainable_parameters, (
"Incorrect non-trainable parameters count")
assert implemented_model.input_shape[1:] == input_shape, (
"Incorrect input shape")
assert implemented_model.output_shape[1:] == output_shape, (
"Incorrect output shape")
del implemented_model


@pytest.mark.parametrize(('min_level, max_level, num_scales, aspect_ratios,'
'anchor_scale, image_size, anchor_count'),
[
(EFFICIENTDETD0, 0, (1, 2, 3, 4, 5)),
(EFFICIENTDETD1, 1, (1, 2, 3, 4, 5)),
(EFFICIENTDETD2, 2, (1, 2, 3, 4, 5)),
(EFFICIENTDETD3, 3, (1, 2, 3, 4, 5)),
(EFFICIENTDETD4, 4, (1, 2, 3, 4, 5)),
(EFFICIENTDETD5, 5, (1, 2, 3, 4, 5)),
(EFFICIENTDETD6, 6, (1, 2, 3, 4, 5)),
(EFFICIENTDETD7, 7, (1, 2, 3, 4, 5))
(3, 7, 3, [1.0, 2.0, 0.5], 4.0, (512, 512),
49104),
(3, 7, 3, [1.0, 2.0, 0.5], 4.0, (640, 640),
76725),
(3, 7, 3, [1.0, 2.0, 0.5], 4.0, (768, 768),
110484),
(3, 7, 3, [1.0, 2.0, 0.5], 4.0, (896, 896),
150381),
(3, 7, 3, [1.0, 2.0, 0.5], 4.0, (1024, 1024),
196416),
(3, 7, 3, [1.0, 2.0, 0.5], 4.0, (1280, 1280),
306900),
(3, 7, 3, [1.0, 2.0, 0.5], 4.0, (1536, 1536),
441936),
])
def test_efficientdet_result(model_input_output_base_path, model,
model_idx, preprocessed_inputs):
for preprocessed_input_idx in preprocessed_inputs:
preprocessed_input_file = model_input_output_base_path + \
'EFFICIENTDETD' + str(model_idx) + '/inputs/test_image_' + \
str(preprocessed_input_idx) + '.npy'
with open(preprocessed_input_file, 'rb') as f:
preprocessed_input = np.load(f)

target_model_output_file = model_input_output_base_path + \
'EFFICIENTDETD' + str(model_idx) + '/outputs/model_output_' + \
str(preprocessed_input_idx) + '.npy'
with open(target_model_output_file, 'rb') as f:
target_model_output = np.load(target_model_output_file)

assert np.all(model()(preprocessed_input).numpy() ==
target_model_output), 'Model result not as expected'
del model
def test_build_prior_boxes(min_level, max_level, num_scales, aspect_ratios,
anchor_scale, image_size, anchor_count):
prior_boxes = build_prior_boxes(
min_level, max_level, num_scales, aspect_ratios, anchor_scale,
image_size)
anchor_x, anchor_y = prior_boxes[:, 0], prior_boxes[:, 1]
anchor_W, anchor_H = prior_boxes[:, 2], prior_boxes[:, 3]
measured_aspect_ratios = set(np.unique(np.round((anchor_W/anchor_H), 2)))
assert np.logical_and(anchor_x >= 0, anchor_x <= 1).all(), (
"Invalid x-coordinates of anchor centre")
assert np.logical_and(anchor_y >= 0, anchor_y <= 1).all(), (
"Invalid y-coordinates of anchor centre")
assert (anchor_W > 0).all(), "Invalid/negative anchor width"
assert (anchor_H > 0).all(), "Invalid/negative anchor height"
assert np.round(np.mean(anchor_x), 2) == 0.5, (
"Anchor boxes asymmetrically distributed along X-direction")
assert np.round(np.mean(anchor_y), 2) == 0.5, (
"Anchor boxes asymmetrically distributed along Y-direction")
assert measured_aspect_ratios == set(aspect_ratios), (
"Anchor aspect ratios not as expected")
assert prior_boxes.shape[0] == anchor_count, (
"Incorrect number of anchor boxes")