import tensorflow as tf
import dataset_util
import numpy as np
model = tf.keras.applications.ResNet50(include_top=True, weights="imagenet")
print(model.summary())
ds = dataset_util.VOC2012_Dataset("dataset/VOC2012", "/train_list.txt", 2)
images, gt_boxes, gt_classes, gt_masks, img_sizes = ds.next_batch()
print(np.argmax(model(images), axis=1))
import sys
sys.exit(0)
last_layer = tf.keras.models.Model(
inputs=model.input,
outputs=[
model.get_layer('conv4_block6_out').output,
model.get_layer('conv4_block5_out').output
])
res = last_layer(images, training=True)
print(res)
if __name__ == "__main__":
anchors = anchor_utils.get_all_anchors_for_c5([16, 16], 128**2,
config.ANCHOR_RATIOS)
backbone2 = backbone.Resnet34_FPN()
model = rpn.RPN(backbone2, 3)
optimizer = tf.keras.optimizers.SGD(lr=0.0005, momentum=0.9, decay=1e-4)
checkpoint = tf.train.Checkpoint(optimizer=optimizer,
net=model,
step=tf.Variable(1))
manager = tf.train.CheckpointManager(checkpoint,
config.WEIGHTS_DIR,
max_to_keep=4)
train_dataset = dataset_util.VOC2012_Dataset("DATASET/VOC2012/VOC2012",
"/train_list.txt", 20)
# train_dataset = dataset_util.VOC2012_Dataset("dataset/VOC2012", "/train_list.txt", 2)
valid_dataset = dataset_util.VOC2012_Dataset("DATASET/VOC2012/VOC2012",
"/valid_list.txt", 20)
# valid_dataset = dataset_util.VOC2012_Dataset("dataset/VOC2012", "/valid_list.txt", 2)
if manager.latest_checkpoint:
print("Restoring...", manager.latest_checkpoint)
images, gt_boxes, gt_classes, gt_masks, img_sizes = train_dataset.next_batch(
)
rpn_classes, rpn_bbox_deltas = anchor_utils.get_rpn_classes_and_bbox_deltas(
len(images), anchors, gt_boxes)
l1, l2 = train_step(
model, optimizer, [images, img_sizes],
[gt_boxes, gt_classes, img_sizes, rpn_classes, rpn_bbox_deltas])
checkpoint.restore(manager.latest_checkpoint).assert_consumed()
import rpn
import backbone
import dataset_util
import anchor_utils
import tensorflow as tf
import config
import mask_rcnn
import numpy as np
import image_util
import metrics
ds = dataset_util.VOC2012_Dataset("dataset/VOC2012", "/valid_list.txt", 2)
#ds = dataset_util.VOC2012_Dataset("dataset/TEST", "/test_list.txt", 2)
anchors = anchor_utils.get_all_anchors(config.IMAGE_SIZE, config.ANCHOR_SCALES,
config.ANCHOR_RATIOS)
backbone2 = backbone.Resnet34_FPN()
rpn2 = rpn.RPN(backbone2, 3)
model = mask_rcnn.Mask_RCNN(rpn2, anchors, len(config.CLASSES))
checkpoint = tf.train.Checkpoint(net=model, step=tf.Variable(1))
manager = tf.train.CheckpointManager(checkpoint,
config.WEIGHTS_DIR,
max_to_keep=4)
if manager.latest_checkpoint:
print("Restoring...", manager.latest_checkpoint)
model([
np.random.rand(1, config.IMAGE_SIZE[0], config.IMAGE_SIZE[1], 3),
np.array([[500, 500]])
],
training=False)
plt.imshow(image/255)
a = np.reshape(softmax[1::3], shape)
a = skt.resize(a, (512, 512))
plt.imshow(a, cmap='Greens', interpolation='nearest', alpha=0.4)
plt.savefig(name + "-2.png")
plt.imshow(image/255)
a = np.reshape(softmax[2::3], shape)
a = skt.resize(a, (512, 512))
plt.imshow(a, cmap='Blues', interpolation='nearest', alpha=0.4)
plt.savefig(name + "-3.png")
if __name__ == "__main__":
#tf.executing_eagerly()
#train_dataset = dataset_util.VOC2012_Dataset("DATASET/VOC2012/VOC2012", "/train_list.txt", 2)
train_dataset = dataset_util.VOC2012_Dataset("dataset/VOC2012", "/train_list.txt", 5)
#valid_dataset = dataset_util.VOC2012_Dataset("DATASET/VOC2012/VOC2012", "/valid_list.txt", 2)
valid_dataset = dataset_util.VOC2012_Dataset("dataset/VOC2012", "/valid_list.txt", 5)
anchors = anchor_utils.get_all_anchors(config.IMAGE_SIZE, config.ANCHOR_SCALES, config.ANCHOR_RATIOS)
backbone2 = backbone.Resnet34_FPN()
rpn2 = rpn.RPN(backbone2, 3)
model = mask_rcnn.Mask_RCNN(rpn2, anchors, len(config.CLASSES))
checkpoint = tf.train.Checkpoint(net=model, step=tf.Variable(1))
manager = tf.train.CheckpointManager(checkpoint, config.WEIGHTS_DIR, max_to_keep=4)
if manager.latest_checkpoint:
print("Restoring...", manager.latest_checkpoint)
images, gt_boxes, gt_classes, gt_masks, img_sizes = valid_dataset.next_batch()
model([images, img_sizes], training=False)
checkpoint.restore(manager.latest_checkpoint)
anchors_scales.append(anchors_scales_i)
anchors_scales = [
[90, 140, 200, 250, 320],
[80, 130, 180, 240, 300],
[85, 150, 210, 260, 340],
[95, 140, 210, 280, 350],
[90, 135, 185, 230, 320]
]
#best [95, 145, 200, 260, 350] 44.07723995880536
anchors = [get_all_anchors((512, 512), scale, [(1, 1), (1, 2), (2, 1)]) for scale in anchors_scales]
sums = np.zeros(len(anchors_scales))
batch_size = 20
ds = dataset_util.VOC2012_Dataset("DATASET/VOC2012/VOC2012", "/all_list.txt", batch_size)
#ds = dataset_util.VOC2012_Dataset("dataset/VOC2012", "/train_list.txt", batch_size)
for j in range(ds.total_batches):
for i in range(len(anchors)):
data1, data2_2, data3, d5, d4 = ds.next_batch()
data2_2, data3_2 = get_rpn_classes_and_bbox_deltas(len(data1), anchors[i], data2_2)
ind = np.where(data2_2 == 1)[0]
sums[i] += len(ind)
print("batch", j, sums, sums / ((j+1) * batch_size))
for i in range(len(anchors_scales)):
print(anchors_scales[i], sums[i] / len(ds.data_names))
print("{}. bigger map in row, exiting".format(bigger_map_in_row))
sys.exit(0)
if __name__ == "__main__":
#tf.executing_eagerly()
anchors = anchor_utils.get_all_anchors(config.IMAGE_SIZE, config.ANCHOR_SCALES, config.ANCHOR_RATIOS)
backbone2 = backbone.Resnet34_FPN()
rpn2 = rpn.RPN(backbone2, 3)
model = mask_rcnn.Mask_RCNN(rpn2, anchors, len(config.CLASSES))
optimizer = tf.keras.optimizers.SGD(lr=0.004, momentum=0.9)
checkpoint = tf.train.Checkpoint(optimizer=optimizer, net=model, step=tf.Variable(1))
manager = tf.train.CheckpointManager(checkpoint, config.WEIGHTS_DIR, max_to_keep=4)
train_dataset = dataset_util.VOC2012_Dataset("DATASET/VOC2012/VOC2012", "/train_list.txt", 12)
td_map = dataset_util.VOC2012_Dataset("DATASET/VOC2012/VOC2012", "/train_list.txt", 4)
vd_map = dataset_util.VOC2012_Dataset("DATASET/VOC2012/VOC2012", "/valid_list.txt", 4)
'''train_dataset = dataset_util.AOLP_Dataset("DATASET/AOLP", "/train_list.txt", 10)
td_map = dataset_util.AOLP_Dataset("DATASET/AOLP", "/train_list.txt", 4)
vd_map = dataset_util.AOLP_Dataset("DATASET/AOLP", "/valid_list.txt", 4)'''
if manager.latest_checkpoint:
print("Restoring...", manager.latest_checkpoint)
images, gt_boxes, gt_classes, gt_masks, img_sizes = train_dataset.next_batch()
rpn_classes, rpn_bbox_deltas = anchor_utils.get_rpn_classes_and_bbox_deltas(len(images), anchors, gt_boxes)
l1, l2, l3, l4, l5 = train_step(model, optimizer, [images, img_sizes], [gt_boxes, gt_classes, gt_masks, img_sizes, rpn_classes, rpn_bbox_deltas])
checkpoint.restore(manager.latest_checkpoint).assert_consumed()
mask_map, box_map = calculate_map(td_map, model)