def test():
config = tf.ConfigProto()
config.gpu_options.allow_growth = True
config.gpu_options.visible_device_list = "0"
with tf.Session(config=config) as sess:
model = create_model(sess, FLAGS.batch_size)
list_of_experiments = glob.glob(os.path.join(".", "log_HG2D", "*"))
p2d_dir = os.path.join(
max(list_of_experiments, key=os.path.getctime),
"predictions") # latest created dir for latest experiment
list_of_preds = glob.glob(os.path.join(p2d_dir, "*"))
print("\n")
for test_fname in list_of_preds:
test_fname = test_fname.split("/")[-1]
print("Converting {} to p3d".format(test_fname))
a = time.time()
encoder_inputs = model.get_test_data(
os.path.join(p2d_dir, test_fname))
encoder_inputs = model.input_normalizer.transform(encoder_inputs)
decoder_outputs = model.test_step(encoder_inputs, sess)
decoder_outputs = model.output_normalizer.inverse_transform(
decoder_outputs)
for i in range(decoder_outputs.shape[0]):
temp = decoder_outputs[i].reshape(17, 3)
temp -= temp[0]
decoder_outputs[i] = temp.reshape(17 * 3)
print("Done in {} s".format(time.time() - a))
experiment = p2d_dir.split("/")[-2]
submissions_dir = os.path.join(
train_dir, "submissions_for_{}".format(experiment))
if not os.path.exists(submissions_dir):
os.makedirs(submissions_dir)
step_nb = test_fname.split(".")[-2].split("_")[-1]
out_file = os.path.join(submissions_dir,
"p3d_step_{}.csv.gz".format(step_nb))
print("saving at: {}".format(out_file))
print("\n\n")
utils.generate_submission_3d(decoder_outputs, out_file)
print("Saving code ...")
submission_files = [
"data.py", "hourglass2D_model.py", "linear_model.py",
"train_hourglass2D.py", "train_simple_baseline.py",
"test_simple_baseline.py", "layers.py", "setup.py", "utils.py",
"vis.py"
]
utils.create_zip_code_files(os.path.join(submissions_dir, "code.zip"),
submission_files)
print("Done.")
help='log storage dir for tensorboard')
opt = parser.parse_args()
with tf.Session() as sess:
# define resnet model
sample = create_test_dataloader(data_root=opt.data_root,
batch_size=opt.batch_size)
with tf.variable_scope('model'):
model = resnet_model.Model()
p3d_out_norm = model(sample['image'], training=False)
p3d_out = unnormalize_pose(p3d_out_norm)
p3d_out = tf.reshape(p3d_out, [-1, 51])
# restore weights
saver = tf.train.Saver()
saver.restore(sess, tf.train.latest_checkpoint(opt.log_dir))
predictions = None
with trange(math.ceil(meta_info.NUM_SAMPLES_TEST / opt.batch_size)) as t:
for i in t:
p3d_out_ = sess.run(p3d_out)
if predictions is None:
predictions = p3d_out_
else:
predictions = np.concatenate([predictions, p3d_out_], axis=0)
generate_submission(predictions, 'submission.csv.gz')
create_zip_code_files('code.zip')
predictions = np.concatenate(
[predictions, p3d_out_value], axis=0)
predictions = predictions.reshape([-1, 51])
print("\nPredictions shape:", predictions.shape)
sys.stdout.flush()
submissions_dir = os.path.join(LOG_PATH, "submissions")
if not os.path.exists(submissions_dir):
os.makedirs(submissions_dir)
utils.generate_submission_3d(
predictions,
os.path.join(
submissions_dir,
"submission_{}.csv.gz".format(global_step_val)))
submission_files = [
"data.py", "hourglass3D_model.py", "test_hourglass3D.py",
"train_hourglass3D.py", "layers.py", "setup.py",
"utils.py", "vis.py"
]
utils.create_zip_code_files(
os.path.join(submissions_dir,
"code_{}.zip".format(global_step_val)),
submission_files)
global_step_val = sess.run(global_step) # get the global step value
saver.save(sess,
os.path.join(CHECKPOINTS_PATH, "model"),
global_step=global_step_val) # save at the end of training
import numpy as np
import os
import math
# Global Attribute
BATCH_SIZE = 1 # needs to be 1 !
DATA_PATH = "/cluster/project/infk/hilliges/lectures/mp19/project2/"
SAVE_PATH= "./submitted_weights"
config = tf.ConfigProto()
config.gpu_options.allow_growth = True
config.gpu_options.visible_device_list = "0"
with tf.Session(config=config) as sess:
testGenerator = DataGenerator(DATA_PATH, batch_size=BATCH_SIZE, name="test")
model=get_model(batch_size=BATCH_SIZE, train=False)
# Load weights into the new model
latest = tf.train.latest_checkpoint(SAVE_PATH)
print(latest)
model.load_weights(latest)
# predict 3d pose
model.compile(optimizer='adam', loss=tf.losses.mean_squared_error)
p3d_out = model.predict_generator(testGenerator, verbose=1)
# compute MPJPE
p3d_out = unnormalize_pose_numpy(p3d_out, 0, 1100*2)
generate_submission(p3d_out,"submission.csv.gz")
create_zip_code_files("code.zip")
all_heatmaps_pred, p3d_pred = model(im, training=False)
# restore weights
print("Restoring latest model from {}\n".format(RESTORE_PATH))
saver = tf.train.Saver()
saver.restore(sess, tf.train.latest_checkpoint(RESTORE_PATH))
predictions = None
with trange(NUM_SAMPLES) as t: # generate predictions for all images
for i in t:
image, p3d_out_value = sess.run([im, p3d_pred])
# image = ((image[0]+1)*128.0).transpose(1,2,0).astype("uint8") # unnormalize, put in channels_last format and cast to uint8
# save_dir = os.path.join(os.getcwd(), "test_samples")
# utils.save_p3d_image(image, None, p3d_out_value[0], save_dir, i+1)
if predictions is None:
predictions = p3d_out_value
else:
predictions = np.concatenate([predictions, p3d_out_value],
axis=0)
predictions = predictions.reshape([-1, 51])
print(predictions.shape)
utils.generate_submission_3d(predictions, "submission.csv.gz")
submission_files = [
"data.py", "hourglass3D_model.py", "test_hourglass3D.py",
"train_hourglass3D.py", "layers.py", "setup.py", "utils.py", "vis.py"
]
utils.create_zip_code_files("code.zip", submission_files)
for f in list_of_files:
preds = pd.read_csv(f, header=0, index_col=0).values
if merged_preds is None:
merged_preds = preds
else:
merged_preds = merged_preds + preds
n += 1
merged_preds = merged_preds / n
print("Done.")
print("Saving merged 3D predictions and code to '{}'".format(OUT_PATH))
if not os.path.exists(OUT_PATH):
os.makedirs(OUT_PATH)
utils.generate_submission_3d(
merged_preds, os.path.join(OUT_PATH, "submission_merged.csv.gz"))
submission_files = [
"data.py", "hourglass2D_model.py", "hourglass3D_model.py",
"linear_model.py", "merge_predictions.py", "test_hourglass2D.py",
"test_hourglass3D.py", "test_simple_baseline.py", "train_hourglass2D.py",
"train_hourglass3D.py", "train_simple_baseline.py", "layers.py",
"setup.py", "utils.py", "vis.py"
]
utils.create_zip_code_files(os.path.join(OUT_PATH, "code.zip"),
submission_files)
print("Done.")