num_predictions = spnet.get_num_predictions(cfg.num_pyramids, cfg.num_levels)
start_lr = 0.001
weights_path = os.path.join(logdir, 'weights_posebaseline_{epoch:03d}.hdf5')
batch_size_mpii = 14
batch_size_ar = 2
"""Load datasets"""
mpii = MpiiSinglePerson(datasetpath('MPII'), dataconf=mpii_dataconf,
poselayout=pa17j3d)
# h36m = Human36M(datasetpath('Human3.6M'), dataconf=human36m_dataconf,
# poselayout=pa17j3d, topology='frames')f
penn_sf = PennAction(datasetpath('Penn_Action'), pennaction_dataconf,
poselayout=pa17j3d, topology='frames', use_gt_bbox=True)
ntu_sf = Ntu(datasetpath('NTU'), ntu_pe_dataconf, poselayout=pa17j3d,
topology='frames', use_gt_bbox=True)
"""Create an object to load data from all datasets."""
data_tr = BatchLoader([mpii, penn_sf, ntu_sf], ['frame'], ['pose'],
TRAIN_MODE, batch_size=[batch_size_mpii, batch_size_ar,
batch_size_ar], num_predictions=num_predictions, shuffle=True)
"""MPII validation samples."""
mpii_val = BatchLoader(mpii, ['frame'], ['pose', 'afmat', 'headsize'],
VALID_MODE, batch_size=mpii.get_length(VALID_MODE), shuffle=True)
printcn(OKBLUE, 'Pre-loading MPII validation data...')
[x_val], [p_val, afmat_val, head_val] = mpii_val[0]
mpii_callback = MpiiEvalCallback(x_val, p_val, afmat_val, head_val,
"E:\\Bachelorarbeit-SS20\\weights\\deephar\\output\\ntu_baseline\\0603",
'weights_posebaseline_{epoch:03d}.hdf5')
batch_size_mpii = 10
batch_size_ar = 2
"""Load datasets"""
mpii = MpiiSinglePerson("D:\\MPII", dataconf=mpii_dataconf, poselayout=pa17j3d)
h36m = Human36M("B:\\Human3.6M",
dataconf=human36m_dataconf,
poselayout=pa17j3d,
topology='frames')
penn_sf = PennAction("D:\\PennAction",
pennaction_dataconf,
poselayout=pa17j3d,
topology='frames',
use_gt_bbox=True)
ntu_sf = Ntu("E:\\Bachelorarbeit-SS20\\datasets\\NTU",
ntu_pe_dataconf,
poselayout=pa17j3d,
topology='frames',
use_gt_bbox=True)
"""Create an object to load data from all datasets."""
data_tr = BatchLoader([mpii, h36m, penn_sf, ntu_sf], ['frame'], ['pose'],
TRAIN_MODE,
batch_size=[
batch_size_mpii, batch_size_mpii, batch_size_ar,
batch_size_ar
],
num_actions,
input_shape,
num_frames,
num_joints,
num_blocks,
pose_dim=2,
pose_net_version='v1',
full_trainable=False)
weights_path = "weights_pennaction_010.h5"
model.load_weights(weights_path)
"""Load PennAction dataset."""
# """
penn_seq = PennAction('datasets/PennAction',
pennaction_dataconf,
poselayout=pa16j2d,
topology='sequences',
use_gt_bbox=use_bbox,
clip_size=num_frames)
penn_te = BatchLoader(penn_seq, ['frame'], ['pennaction'],
TEST_MODE,
batch_size=1,
shuffle=False)
# =================================================================
import numpy as np
import json
import matplotlib.pyplot as plt
frame_list = penn_seq.get_clip_index(0, TEST_MODE, subsamples=[2])
with open('datasets/PennAction/penn_pred_bboxes_16f.json', 'r') as fid:
bboxes_data = json.load(fid)
num_frames,
num_joints,
num_blocks,
pose_dim=2,
pose_net_version='v1',
full_trainable=False)
"""Load pre-trained model."""
weights_path = get_file(weights_file,
TF_WEIGHTS_PATH,
md5_hash=md5_hash,
cache_subdir='models')
model.load_weights(weights_path)
"""Load PennAction dataset."""
penn_seq = PennAction('datasets/PennAction',
pennaction_dataconf,
poselayout=pa16j2d,
topology='sequences',
use_gt_bbox=use_bbox,
clip_size=num_frames)
penn_te = BatchLoader(penn_seq, ['frame'], ['pennaction'],
TEST_MODE,
batch_size=1,
shuffle=False)
printcn(OKGREEN, 'Evaluation on PennAction multi-clip using predicted bboxes')
eval_multiclip_dataset(
model,
penn_seq,
bboxes_file='datasets/PennAction/penn_pred_bboxes_16f.json',
logdir=logdir)
num_frames = 8
num_clips = 250
cfg = ModelConfig((num_frames,) + pennaction_dataconf.input_shape, pa16j2d,
num_actions=[15], num_pyramids=6, action_pyramids=[1, 2, 3, 4, 5, 6],
num_levels=4, pose_replica=True,
num_pose_features=160, num_visual_features=160)
num_predictions = spnet.get_num_predictions(cfg.num_pyramids, cfg.num_levels)
num_action_predictions = \
spnet.get_num_predictions(len(cfg.action_pyramids), cfg.num_levels)
"""Load PennAction"""
penn_seq = PennAction(datasetpath('Penn_Action'), pennaction_dataconf,
poselayout=pa16j2d, topology='sequences', use_gt_bbox=False,
pred_bboxes_file='pred_bboxes_penn.json', clip_size=num_frames)
"""Build the full model"""
full_model = spnet.build(cfg)
"""Load pre-trained weights from pose estimation and copy replica layers."""
full_model.load_weights(
'output/penn_multimodel_trial-07-full_2e9fa5a/weights_mpii+penn_ar_028.hdf5',
by_name=True)
"""Pre-load some samples from PennAction."""
penn_te = BatchLoader(penn_seq, ['frame'], ['pennaction'], TEST_MODE,
batch_size=num_clips, shuffle=False)
[x], [y] = penn_te[0]
# Check file with bounding boxes
penn_data_path = os.getcwd() + '/datasets/PennAction'
penn_bbox_file = 'penn_pred_bboxes_multitask.json'
if os.path.isfile(os.path.join(penn_data_path, penn_bbox_file)) == False:
logger.debug(
f'Error: file {penn_bbox_file} not found in {penn_data_path}!')
logger.debug(
f'\nPlease download it from https://drive.google.com/file/d/1qXpEKF0d9KxmQdd2_QSIA1c3WGj1D3Y3/view?usp=sharing'
)
sys.stdout.flush()
sys.exit()
penn_seq = PennAction(penn_data_path,
pennaction_dataconf,
poselayout=pa16j2d,
topology='sequences',
use_gt_bbox=False,
pred_bboxes_file='penn_pred_bboxes_multitask.json',
clip_size=num_frames)
logger.info("PENN ACTION Loaded")
logger.info("Build FULL MODEL")
"""Build the full model"""
full_model = spnet.build(cfg)
weights_file = os.getcwd() + '/weights/weights_mpii+penn_ar_028.hdf5'
if os.path.isfile(weights_file) == False:
logger.debug(f'Error: file {weights_file} not found!')
logger.debug(
f'\nPlease download it from https://drive.google.com/file/d/106yIhqNN-TrI34SX81q2xbU-NczcQj6I/view?usp=sharing'
from datasetpath import datasetpath
from generic import get_bbox_from_poses
logdir = './'
if len(sys.argv) > 1:
logdir = sys.argv[1]
mkdir(logdir)
sys.stdout = open(str(logdir) + '/log.txt', 'w')
cfg = ModelConfig(dconf.input_shape, pa16j2d, num_pyramids=8, num_levels=4)
"""Load dataset"""
dpath = datasetpath('Penn_Action')
penn = PennAction(dpath,
dconf,
poselayout=pa16j2d,
topology='frames',
use_gt_bbox=False)
"""Build and compile the network."""
model = spnet.build(cfg)
model.load_weights(
'output/mpii_spnet_51b_741a720/weights_mpii_spnet_8b4l_050.hdf5')
"""Squeeze the model for only one output."""
model = Model(model.input, model.outputs[-1])
def predict_frame_bboxes(mode):
bboxes = {}
num_samples = penn.get_length(mode)
num_predictions = spnet.get_num_predictions(cfg.num_pyramids, cfg.num_levels)
num_action_predictions = \
spnet.get_num_predictions(len(cfg.action_pyramids), cfg.num_levels)
start_lr = 0.001
action_weight = 0.01
batch_size_mpii = int(0.8 * num_frames)
# batch_size_penn = num_frames - batch_size_mpii
batch_size_penn = num_frames
batch_clips = 4 # 8/4
"""Load datasets"""
mpii = MpiiSinglePerson("D:\\MPII", dataconf=mpii_dataconf, poselayout=pa16j2d)
penn_sf = PennAction("D:\\PennAction",
pennaction_pe_dataconf,
poselayout=pa16j2d,
topology='frames',
use_gt_bbox=True)
penn_seq = PennAction("D:\\PennAction",
pennaction_dataconf,
poselayout=pa16j2d,
topology='sequences',
use_gt_bbox=True,
clip_size=num_frames)
# pe_data_tr = BatchLoader([mpii, penn_sf], ['frame'], ['pose'], TRAIN_MODE,
pe_data_tr = BatchLoader(
[mpii],
['frame'],
['pose'],
#from datasetpath import datasetpath
from generic import get_bbox_from_poses
logdir = './'
if len(sys.argv) > 1:
logdir = sys.argv[1]
mkdir(logdir)
sys.stdout = open(str(logdir) + '/log.txt', 'w')
cfg = ModelConfig(dconf.input_shape, pa16j2d, num_pyramids=8, num_levels=4)
"""Load dataset"""
#dpath = datasetpath('Penn_Action')
penn = PennAction('E:\\Bachelorarbeit-SS20\\datasets\\PennAction',
dconf,
poselayout=pa16j2d,
topology='frames',
use_gt_bbox=False)
"""Build and compile the network."""
model = spnet.build(cfg)
model.load_weights(
'output/mpii_spnet_51b_741a720/weights_mpii_spnet_8b4l_050.hdf5')
"""Squeeze the model for only one output."""
model = Model(model.input, model.outputs[-1])
def predict_frame_bboxes(mode):
bboxes = {}
num_samples = penn.get_length(mode)
#num_frames = 16
use_bbox = False
pred_bboxes_file = 'penn_pred_bboxes_16f.json'
num_blocks = 4
batch_size = 2
input_shape = pennaction_dataconf.input_shape
num_joints = 16
num_actions = 15
"""Load datasets"""
#mpii = MpiiSinglePerson('E:\Bachelorarbeit-SS20\datasets\MPII', dataconf=mpii_dataconf,
# poselayout=pa16j2d)
"""Load PennAction dataset."""
penn_seq = PennAction('E:\Bachelorarbeit-SS20\datasets\PennAction',
pennaction_dataconf,
poselayout=pa16j2d,
topology='sequences',
use_gt_bbox=use_bbox,
pred_bboxes_file=pred_bboxes_file,
clip_size=num_frames)
"""Build the full model"""
full_model = spnet.build(cfg)
"""Load pre-trained weights from pose estimation and copy replica layers."""
full_model.load_weights('output/weights_AR_merge_ep074_26-10-17.h5',
by_name=True)
"""This call splits the model into its parts: pose estimation and action
recognition, so we can evaluate each part separately on its respective datasets.
"""
models = split_model(full_model,
cfg,
interlaced=False,
model_names=['2DPose', '2DAction'])
