luni/cvpr_2023
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

pyramid pipeline ready

Browse Source
This commit is contained in:
luni
2023-05-22 00:13:39 +02:00
parent 8e0d061104
commit 52e461ca47
6 changed files with 938 additions and 793 deletions
Download Patch File Download Diff File Expand all files Collapse all files
scripts/cvae_pyramid.py
+279 -302
View File
@@ -1,65 +1,67 @@
import os
import sys
from configs.config_abl_pyr_32_skirt import config
sys.path.append(os.path.dirname(os.path.abspath(__file__)) + '/..')
import importlib
import argparse
parser = argparse.ArgumentParser(description='')
parser.add_argument('--name', type=str)
parser.add_argument('--config', type=str)
parser.add_argument('--debug', action='store_true')
args = vars(parser.parse_args())
config_module = importlib.import_module(args['config'])
config = config_module.config
if args['debug']:
from configs.pyr_32 import config
os.environ["CUDA_DEVICE_ORDER"]="PCI_BUS_ID"
os.environ["CUDA_VISIBLE_DEVICES"] = config['gpu']
import tensorflow as tf
gpus = tf.config.experimental.list_physical_devices('GPU')
if gpus:
try:
# Currently, memory growth needs to be the same across GPUs
for gpu in gpus:
tf.config.experimental.set_memory_growth(gpu, True)
logical_gpus = tf.config.experimental.list_logical_devices('GPU')
print(len(gpus), "Physical GPUs,", len(logical_gpus), "Logical GPUs")
except RuntimeError as e:
# Memory growth must be set before GPUs have been initialized
print(e)
if args['debug']:
os.environ["CUDA_LAUNCH_BLOCKING"] = '1'
import time
import tqdm
import argparse
import wandb
import tqdm
import random
import numpy as np
import scipy as sp
from custom_utils.custom_utils import UV2Mesh, get_summary_html, maskFacesBatch, _maskFaces, get_summary_str, plot_3d_collision, group_inputs_for_levels
import plotly.graph_objects as go
import plotly.express as ex
from plotly.subplots import make_subplots
from PIL import Image
if config['is_skirt']:
from losses import edge_loss_skirt as edge_loss
from losses.losses import edge_loss_skirt as edge_loss
else:
from losses import edge_loss
from losses.losses import edge_loss
from losses import uv_mask_loss, bend_loss, kl_loss, collision_ratio, normal_loss
# from losses import collision_penalty as collision_penalty
from losses import collision_loss as collision_penalty
from custom_utils.custom_utils import batch_to_dict, load_model
from dataset import DatasetFactory, StaticDataset
from custom_utils.smpl import SMPL
from models.pyramid_model_optimized import Pyramid
from postprocessing import fix_collision
import torch
import torch.nn as nn
import torch.optim as optim
import torch.nn.functional as F
from data.dataset import Dataset
from utils.checkpoints import CheckpointSaver, load_model
from utils.utils import InfiniteDataLoader as DataLoader
from utils.postprocess import fix_collision
from losses.losses import collision_loss, normal_loss, kl_loss
from pytorch_msssim import MS_SSIM
from custom_utils.custom_utils import UV2Mesh, Accumulator, plot_3d, get_summary_html, maskFacesBatch, get_summary_str, plot_3d_collision, group_inputs_for_levels
from smpl.smpl_torch import SMPL
from plotly.subplots import make_subplots
from models.pyramid_model import Pyramid
def init_wandb():
run = wandb.init(project='ablation', name=config['name'], notes=config['notes'], dir='/wandb', tags=config['tags'])
run = wandb.init(project='iccv', name=config['name'], notes=config['notes'], dir='/data-local/127/temp', tags=config['tags'])
artifact = wandb.Artifact('code', type='code')
artifact.add_file('cvae_pyramid.py')
artifact.add_file('models/cvae_model.py')
artifact.add_file('models/pyramid_model_optimized.py')
artifact.add_file('models/pyramid_level.py')
artifact.add_file('models/upscaler.py')
artifact.add_file('losses.py')
artifact.add_file('scripts/cvae_pyramid.py')
artifact.add_file('losses/losses.py')
artifact.add_file('models/base.py')
artifact.add_file('models/cond_encoder.py')
artifact.add_file('dataset.py')
artifact.add_file('convnext/models/convnext_tf.py')
config_file_names = [key for key in sys.modules.keys() if key.startswith('configs')]
artifact.add_file('models/convnext.py')
artifact.add_file('models/pyramid_model.py')
artifact.add_file('models/pyramid_level.py')
artifact.add_file('data/dataset.py')
config_file_names = [key for key in sys.modules.keys() if key.startswith('configs.')]
for config_file_name in config_file_names:
file_name = '/'.join(config_file_name.split('.')) + '.py'
if os.path.exists(file_name):
@@ -210,86 +212,12 @@ def plot_3d(mesh_gt, mesh_pred_unposed_levels, faces, mesh_posed, collision, col
return fig
# @tf.function
def train_step(batch):
model.train()
# with torch.autocast(device_type='cuda', dtype=torch.float16, enabled=use_amp):
levels_input, conds = group_inputs_for_levels(batch, config)
with tf.GradientTape() as tape:
out_levels = model([levels_input, conds], training=True)
if config['use_skinning']:
mesh_orig, uv_orig = batch['mesh_unposed'], batch['uv_unposed']
else:
mesh_orig, uv_orig = batch['mesh_posed'], batch['uv_posed']
output = dict()
reconstr_loss_total = 0
kl_total = 0
#for i, res in enumerate(config['pyramid']['resolutions']):
i = -1
res = config['pyramid']['resolutions'][-1]
res_modifier = '_' + str(res) if res != 512 else ''
pred, mean, logvar = out_levels[i]
mesh_pred = uv2mesh(pred)
if config['use_skinning']:
shape = batch['pose_shape'][:, 72:]
gender = tf.cast(tf.greater_equal(shape[:, 10] , 0.5), tf.int32)
shape_offsets = tf.stack([tf.gather(tf.tensordot(shape[i,:10][tf.newaxis, :], smpl.shapedirs[gender[i]], axes=[[1], [2]])[0,:,:], smpl.to_keep, axis=0) for i in range(shape.shape[0])])
mesh_pred_shaped = mesh_pred + shape_offsets * batch['mesh_mask'][:,:,tf.newaxis]
mesh_pred_posed = smpl.run_garment(batch['pose_shape'][:, :72], batch['pose_shape'][:, 72:], mesh_pred_shaped)
else:
mesh_pred_posed = mesh_pred
# is the output already masked?
e_loss = tf.reduce_mean(edge_loss(mesh_pred, mesh_orig, batch['mesh_mask']))
c_loss, c_ratio = collision_penalty(mesh_pred_posed, batch['mesh_body_posed'], batch['mesh_mask'])
c_loss, c_ratio = tf.reduce_mean(c_loss), tf.reduce_mean(c_ratio)
l2_loss = tf.reduce_mean(1e3 * tf.reduce_mean(tf.reduce_sum(((uv_orig - pred) ** 2), axis=-1), axis=[1, 2]))
# ms_ssim = tf.reduce_mean(tf.image.ssim_multiscale(pred + 5, batch['uv_unposed'] + 5, max_val=10, power_factors=(0.5, 0.2, 0.1, 0.1, 0.1)))
# ms_ssim = tf.reduce_mean(tf.image.ssim(pred + 1, batch['uv_unposed'] + 1, max_val=3))
l2_mesh_loss = tf.reduce_mean(1e3 * tf.reduce_mean(tf.reduce_sum(batch['mesh_mask'][:,:,tf.newaxis] * (batch['mesh_posed'] - mesh_pred_posed) ** 2, axis=-1), axis=-1))
n_loss = normal_loss(mesh_pred_posed, batch['mesh_posed'], batch['mesh_mask'], is_skirt=config['is_skirt'])
kl = kl_loss(mean, logvar)
if config['use_extra_losses']:
reconstr_loss_total = reconstr_loss_total + config['pyramid']['loss_weights'][i] * (l2_loss + l2_mesh_loss)
if not config['is_cape']:
reconstr_loss_total = reconstr_loss_total + config['pyramid']['loss_weights'][i] * (e_loss + c_loss) * config['reg_loss_coeff']
else:
reconstr_loss_total = reconstr_loss_total + config['pyramid']['loss_weights'][i] * (l2_loss)
# if config['use_norm_loss'] and step > 2000:
# reconstr_loss_total = reconstr_loss_total + n_loss * config['reg_loss_coeff']
kl_total += kl
mean = tf.reduce_mean(mean)
logvar = tf.reduce_mean(logvar)
metric = tf.reduce_sum(tf.math.sqrt(tf.reduce_sum(batch['mesh_mask'][:,:,tf.newaxis] * (mesh_orig - mesh_pred) ** 2, axis=-1))) / tf.reduce_sum(batch['mesh_mask'])
output = { **output, **{ 'train/edge_loss' + res_modifier: e_loss, 'train/collision_loss' + res_modifier: c_loss,'train/euclid_error' + res_modifier: metric,
'train/mean' + res_modifier: mean, 'train/logvar' + res_modifier: logvar, 'train/kl_loss' + res_modifier: kl,
'train/var' + res_modifier: np.e ** logvar, 'train/l2_loss' + res_modifier: l2_loss, 'train/normal_loss': n_loss,
'train/euclid_error': metric, 'train/l2_mesh_loss': l2_mesh_loss } } # 'train/ms_ssim' + res_modifier: ms_ssim,
loss = reconstr_loss_total + balancing_coeff * kl_total
gradients = tape.gradient(loss, model.trainable_variables)
optimizer.apply_gradients(zip(gradients, model.trainable_variables))
return { **{ 'train/loss': loss, 'train/reconstr_loss_total': reconstr_loss_total, 'train/kl_total': kl_total }, **output }
@tf.function
def valid_step(batch):
levels_input, conds = group_inputs_for_levels(batch, config)
out_levels = model([levels_input, conds], training=False)
out_levels = model([levels_input, conds])
if config['use_skinning']:
mesh_orig, uv_orig = batch['mesh_unposed'], batch['uv_unposed']
@@ -310,23 +238,24 @@ def valid_step(batch):
if config['use_skinning']:
shape = batch['pose_shape'][:, 72:]
gender = tf.cast(tf.greater_equal(shape[:, 10] , 0.5), tf.int32)
shape_offsets = tf.stack([tf.gather(tf.tensordot(shape[i,:10][tf.newaxis, :], smpl.shapedirs[gender[i]], axes=[[1], [2]])[0,:,:], smpl.to_keep, axis=0) for i in range(shape.shape[0])])
gender = torch.greater_equal(shape[:, 10] , 0.5).to(dtype=torch.int64)
shape_offsets = torch.stack([torch.tensordot(shape[i,:10][None, :], smpl.shapedirs[gender[i]], dims=[[1], [2]])[0,smpl.to_keep,:] for i in range(shape.shape[0])])
mesh_pred_shaped = mesh_pred + shape_offsets * batch['mesh_mask'][:,:,tf.newaxis]
mesh_pred_posed = smpl.run_garment(batch['pose_shape'][:, :72], batch['pose_shape'][:, 72:], mesh_pred_shaped)
mesh_pred_shaped = mesh_pred + shape_offsets * batch['mesh_mask'][:,:,None]
mesh_pred_posed = smpl(batch['pose_shape'][:, :72], batch['pose_shape'][:, 72:], mesh_pred_shaped)[0]
else:
mesh_pred_posed = mesh_pred
# is the output already masked?
e_loss = tf.reduce_mean(edge_loss(mesh_pred, mesh_orig, batch['mesh_mask']))
c_loss, c_ratio = collision_penalty(mesh_pred_posed, batch['mesh_body_posed'], batch['mesh_mask'])
c_loss, c_ratio = tf.reduce_mean(c_loss), tf.reduce_mean(c_ratio)
l2_loss = tf.reduce_mean(1e3 * tf.reduce_mean(tf.reduce_sum(((uv_orig - pred) ** 2), axis=-1), axis=[1, 2]))
# ms_ssim = tf.reduce_mean(tf.image.ssim_multiscale(pred + 5, batch['uv_unposed'] + 5, max_val=10, power_factors=(0.5, 0.2, 0.1, 0.1, 0.1)))
# ms_ssim = tf.reduce_mean(tf.image.ssim(pred + 1, batch['uv_unposed'] + 1, max_val=3))
l2_mesh_loss = tf.reduce_mean(1e3 * tf.reduce_mean(tf.reduce_sum(batch['mesh_mask'][:,:,tf.newaxis] * (batch['mesh_posed'] - mesh_pred_posed) ** 2, axis=-1), axis=-1))
n_loss = normal_loss(mesh_pred_posed, batch['mesh_posed'], batch['mesh_mask'], is_skirt=config['is_skirt'])
# is the output already masked?
# ms_ssim = torch.mean(tf.image.ssim_multiscale(pred + 5, batch['uv_unposed'] + 5, max_val=10, power_factors=(0.5, 0.2, 0.1, 0.1, 0.1)))
# ms_ssim = torch.mean(tf.image.ssim(pred + 1, batch['uv_unposed'] + 1, max_val=3))
e_loss = torch.mean(edge_loss(mesh_pred, mesh_orig, batch['mesh_mask']))
c_loss, c_ratio = collision_loss(mesh_pred_posed, batch['mesh_body_posed'], batch['mesh_mask'])
c_loss, c_ratio = torch.mean(c_loss), torch.mean(c_ratio)
l2_loss = torch.mean(1e3 * torch.mean(torch.sum(((uv_orig - pred) ** 2), dim=-1), dim=[1, 2]))
l2_mesh_loss = torch.mean(1e3 * torch.mean(torch.sum(batch['mesh_mask'][:,:,None] * (batch['mesh_posed'] - mesh_pred_posed) ** 2, dim=-1), dim=-1))
n_loss = normal_loss(mesh_pred_posed, batch['mesh_posed'], batch['mesh_mask'], is_skirt=config['is_skirt'], is_smpl_orig=config['is_cape'])
kl = kl_loss(mean, logvar)
if config['use_extra_losses']:
@@ -335,35 +264,121 @@ def valid_step(batch):
reconstr_loss_total = reconstr_loss_total + config['pyramid']['loss_weights'][i] * (e_loss + c_loss) * config['reg_loss_coeff']
else:
reconstr_loss_total = reconstr_loss_total + config['pyramid']['loss_weights'][i] * (l2_loss)
if config['use_norm_loss'] and step > 2000:
reconstr_loss_total = reconstr_loss_total + n_loss * config['reg_loss_coeff']
# if config['use_norm_loss'] and step > 2000:
# reconstr_loss_total = reconstr_loss_total + n_loss * config['reg_loss_coeff']
kl_total += kl
loss = reconstr_loss_total + g_balancing_coeff * kl_total
mean = tf.reduce_mean(mean)
logvar = tf.reduce_mean(logvar)
# scaler.scale(loss).backward()
# scaler.step(optimizer)
# scaler.update()
# optimizer.zero_grad()
metric = tf.reduce_sum(tf.math.sqrt(tf.reduce_sum(batch['mesh_mask'][:,:,tf.newaxis] * (mesh_orig - mesh_pred) ** 2, axis=-1))) / tf.reduce_sum(batch['mesh_mask'])
optimizer.zero_grad()
loss.backward()
optimizer.step()
model.eval()
with torch.inference_mode():
mean = torch.mean(mean)
logvar = torch.mean(logvar)
ms_ssim = torch.mean(_MSSSIM(pred + 1, batch['uv_static'] + 1))
metric = torch.sum(torch.sqrt(torch.sum(batch['mesh_mask'][:,:,None] * (mesh_orig - mesh_pred.detach()) ** 2, axis=-1))) / torch.sum(batch['mesh_mask'])
output = { **output, **{ 'train/edge_loss' + res_modifier: e_loss, 'train/collision_loss' + res_modifier: c_loss,'train/euclid_error' + res_modifier: metric,
'train/mean' + res_modifier: mean, 'train/logvar' + res_modifier: logvar, 'train/kl_loss' + res_modifier: kl,
'train/var' + res_modifier: np.e ** logvar, 'train/l2_loss' + res_modifier: l2_loss, 'train/normal_loss': n_loss,
'train/euclid_error': metric, 'train/l2_mesh_loss': l2_mesh_loss, 'train/ms_ssim' + res_modifier: ms_ssim, } } #
return { **{ 'train/loss': loss, 'train/reconstr_loss_total': reconstr_loss_total, 'train/kl_total': kl_total }, **output }
def valid_step(batch):
model.eval()
with torch.inference_mode():
levels_input, conds = group_inputs_for_levels(batch, config)
out_levels = model([levels_input, conds])
if config['use_skinning']:
mesh_orig, uv_orig = batch['mesh_unposed'], batch['uv_unposed']
else:
mesh_orig, uv_orig = batch['mesh_posed'], batch['uv_posed']
output = dict()
reconstr_loss_total = 0
kl_total = 0
#for i, res in enumerate(config['pyramid']['resolutions']):
i = -1
res = config['pyramid']['resolutions'][-1]
res_modifier = '_' + str(res) if res != 512 else ''
pred, mean, logvar = out_levels[i]
mesh_pred = uv2mesh(pred)
if config['use_skinning']:
shape = batch['pose_shape'][:, 72:]
gender = torch.greater_equal(shape[:, 10] , 0.5).to(dtype=torch.int64)
shape_offsets = torch.stack([torch.tensordot(shape[i,:10][None, :], smpl.shapedirs[gender[i]], dims=[[1], [2]])[0,smpl.to_keep,:] for i in range(shape.shape[0])])
mesh_pred_shaped = mesh_pred + shape_offsets * batch['mesh_mask'][:,:,None]
mesh_pred_posed = smpl(batch['pose_shape'][:, :72], batch['pose_shape'][:, 72:], mesh_pred_shaped)[0]
else:
mesh_pred_posed = mesh_pred
# is the output already masked?
# ms_ssim = torch.mean(tf.image.ssim_multiscale(pred + 5, batch['uv_unposed'] + 5, max_val=10, power_factors=(0.5, 0.2, 0.1, 0.1, 0.1)))
# ms_ssim = torch.mean(tf.image.ssim(pred + 1, batch['uv_unposed'] + 1, max_val=3))
e_loss = torch.mean(edge_loss(mesh_pred, mesh_orig, batch['mesh_mask']))
c_loss, c_ratio = collision_loss(mesh_pred_posed, batch['mesh_body_posed'], batch['mesh_mask'])
c_loss, c_ratio = torch.mean(c_loss), torch.mean(c_ratio)
l2_loss = torch.mean(1e3 * torch.mean(torch.sum(((uv_orig - pred) ** 2), dim=-1), dim=[1, 2]))
l2_mesh_loss = torch.mean(1e3 * torch.mean(torch.sum(batch['mesh_mask'][:,:,None] * (batch['mesh_posed'] - mesh_pred_posed) ** 2, dim=-1), dim=-1))
n_loss = normal_loss(mesh_pred_posed, batch['mesh_posed'], batch['mesh_mask'], is_skirt=config['is_skirt'], is_smpl_orig=config['is_cape'])
kl = kl_loss(mean, logvar)
if config['use_extra_losses']:
reconstr_loss_total = reconstr_loss_total + config['pyramid']['loss_weights'][i] * (l2_loss + l2_mesh_loss)
if not config['is_cape']:
reconstr_loss_total = reconstr_loss_total + config['pyramid']['loss_weights'][i] * (e_loss + c_loss) * config['reg_loss_coeff']
else:
reconstr_loss_total = reconstr_loss_total + config['pyramid']['loss_weights'][i] * (l2_loss)
if config['use_norm_loss'] and step > 2000:
reconstr_loss_total = reconstr_loss_total + n_loss * config['reg_loss_coeff']
kl_total += kl
loss = reconstr_loss_total + g_balancing_coeff * kl_total
mean = torch.mean(mean)
logvar = torch.mean(logvar)
ms_ssim = torch.mean(_MSSSIM(pred + 1, batch['uv_static'] + 1))
metric = torch.sum(torch.sqrt(torch.sum(batch['mesh_mask'][:,:,None] * (mesh_orig - mesh_pred.detach()) ** 2, axis=-1))) / torch.sum(batch['mesh_mask'])
output = { **output, **{ 'valid/edge_loss' + res_modifier: e_loss, 'valid/collision_loss' + res_modifier: c_loss,'valid/euclid_error' + res_modifier: metric,
'valid/mean' + res_modifier: mean, 'valid/logvar' + res_modifier: logvar, 'valid/kl_loss' + res_modifier: kl,
'valid/var' + res_modifier: np.e ** logvar, 'valid/l2_loss' + res_modifier: l2_loss, 'valid/normal_loss': n_loss,
'valid/euclid_error': metric, 'valid/l2_mesh_loss': l2_mesh_loss, 'valid/ms_ssim' + res_modifier: ms_ssim, } } #
output = { **output, **{ 'valid/edge_loss' + res_modifier: e_loss, 'valid/collision_loss' + res_modifier: c_loss,'valid/euclid_error' + res_modifier: metric,
'valid/mean' + res_modifier: mean, 'valid/logvar' + res_modifier: logvar, 'valid/kl_loss' + res_modifier: kl,
'valid/var' + res_modifier: np.e ** logvar, 'valid/l2_loss' + res_modifier: l2_loss, 'valid/normal_loss': n_loss,
'valid/euclid_error': metric, 'valid/l2_mesh_loss': l2_mesh_loss } } # 'valid/ms_ssim' + res_modifier: ms_ssim,
loss = reconstr_loss_total + balancing_coeff * kl_total
return { **{ 'valid/loss': loss, 'valid/reconstr_loss_total': reconstr_loss_total, 'valid/kl_total': kl_total }, **output }
@tf.function
def test_step(batch):
levels_input, conds = group_inputs_for_levels(batch, config)
out_levels = model([levels_input, conds], training=False)
shape = batch['pose_shape'][:, 72:]
gender = tf.cast(tf.greater_equal(shape[:, 10] , 0.5), tf.int32)
shape_offsets = tf.stack([tf.gather(tf.tensordot(shape[i,:10][tf.newaxis, :], smpl.shapedirs[gender[i]], axes=[[1], [2]])[0,:,:], smpl.to_keep, axis=0) for i in range(shape.shape[0])])
shape_offsets = tf.stack([tf.gather(tf.tensordot(shape[i,:10][None, :], smpl.shapedirs[gender[i]], axes=[[1], [2]])[0,:,:], smpl.to_keep, dim=0) for i in range(shape.shape[0])])
mesh_orig = batch['mesh_unposed']
output = dict()
@@ -377,18 +392,18 @@ def test_step(batch):
pred, mean, logvar = out_levels[i]
mesh_pred = uv2mesh(pred)
mesh_pred_shaped = mesh_pred + shape_offsets * batch['mesh_mask'][:,:,tf.newaxis]
mesh_pred_shaped = mesh_pred + shape_offsets * batch['mesh_mask'][:,:,None]
mesh_pred_posed = smpl.run_garment(batch['pose_shape'][:, :72], batch['pose_shape'][:, 72:], mesh_pred_shaped)
# is the output already masked?
e_loss = tf.reduce_mean(edge_loss(mesh_pred, mesh_orig, batch['mesh_mask']))
e_loss = torch.mean(edge_loss(mesh_pred, mesh_orig, batch['mesh_mask']))
c_loss, c_ratio = collision_penalty(mesh_pred_posed, batch['mesh_body_posed'], batch['mesh_mask'])
c_loss, c_ratio = tf.reduce_mean(c_loss), tf.reduce_mean(c_ratio)
l2_loss = tf.reduce_mean(1e3 * tf.reduce_mean(tf.reduce_sum(((batch['uv_unposed'] - pred) ** 2), axis=-1), axis=[1, 2]))
l2_mesh_loss = tf.reduce_mean(1e3 * tf.reduce_mean(tf.reduce_sum(batch['mesh_mask'][:,:,tf.newaxis] * (mesh_orig - mesh_pred) ** 2, axis=-1), axis=-1))
# ms_ssim = tf.reduce_mean(tf.image.ssim_multiscale(pred + 5, batch['uv_unposed'] + 5, max_val=10, power_factors=(0.5, 0.2, 0.1, 0.1, 0.1)))
# ms_ssim = tf.reduce_mean(tf.image.ssim(pred + 1, batch['uv_unposed'] + 1, max_val=3))
c_loss, c_ratio = torch.mean(c_loss), torch.mean(c_ratio)
l2_loss = torch.mean(1e3 * torch.mean(torch.sum(((batch['uv_unposed'] - pred) ** 2), dim=-1), dim=[1, 2]))
l2_mesh_loss = torch.mean(1e3 * torch.mean(torch.sum(batch['mesh_mask'][:,:,None] * (mesh_orig - mesh_pred) ** 2, dim=-1), dim=-1))
# ms_ssim = torch.mean(tf.image.ssim_multiscale(pred + 5, batch['uv_unposed'] + 5, max_val=10, power_factors=(0.5, 0.2, 0.1, 0.1, 0.1)))
# ms_ssim = torch.mean(tf.image.ssim(pred + 1, batch['uv_unposed'] + 1, max_val=3))
kl = kl_loss(mean, logvar)
reconstr_loss_total = reconstr_loss_total + config['pyramid']['loss_weights'][i] * (l2_loss + l2_mesh_loss)
@@ -397,11 +412,11 @@ def test_step(batch):
# reconstr_loss_total += config['pyramid']['loss_weights'][i] * (l2_loss + (e_loss + c_loss) * config['reg_loss_coeff'])
kl_total += kl
mean = tf.reduce_mean(mean)
logvar = tf.reduce_mean(logvar)
mean = torch.mean(mean)
logvar = torch.mean(logvar)
metric = tf.reduce_sum(tf.math.sqrt(tf.reduce_sum(batch['mesh_mask'][:,:,tf.newaxis] * (mesh_orig - mesh_pred) ** 2, axis=-1))) / tf.reduce_sum(batch['mesh_mask'])
euclid_error_per_garment = tf.reduce_sum(tf.math.sqrt(tf.reduce_sum(batch['mesh_mask'][:,:,tf.newaxis] * (mesh_orig - mesh_pred) ** 2, axis=-1)), axis=1) / tf.reduce_sum(batch['mesh_mask'], axis=1)
metric = torch.sum(tf.math.sqrt(torch.sum(batch['mesh_mask'][:,:,None] * (mesh_orig - mesh_pred) ** 2, dim=-1))) / torch.sum(batch['mesh_mask'])
euclid_error_per_garment = torch.sum(tf.math.sqrt(torch.sum(batch['mesh_mask'][:,:,None] * (mesh_orig - mesh_pred) ** 2, dim=-1)), dim=1) / torch.sum(batch['mesh_mask'], dim=1)
output = { **output, **{ 'test/edge_loss' + res_modifier: e_loss, 'test/collision_loss' + res_modifier: c_loss,'test/euclid_error' + res_modifier: metric,
'test/mean' + res_modifier: mean, 'test/logvar' + res_modifier: logvar, 'test/kl_loss' + res_modifier: kl,
@@ -416,50 +431,52 @@ def test_step(batch):
def plot_everything(batch: dict, type: str):
if config['use_skinning']:
mesh_orig, uv_orig = batch['mesh_unposed'].numpy(), batch['uv_unposed'].numpy()
mesh_orig, uv_orig = batch['mesh_unposed'].cpu().numpy(), batch['uv_unposed']
else:
mesh_orig, uv_orig = batch['mesh_posed'].numpy(), batch['uv_posed'].numpy()
mesh_orig, uv_orig = batch['mesh_posed'].cpu().numpy(), batch['uv_posed']
levels_input = group_inputs_for_levels(batch, config)
output_levels = model(levels_input, training=False)
model.eval()
with torch.inference_mode():
levels_input = group_inputs_for_levels(batch, config)
output_levels = model(levels_input)
mesh_pred_levels = [uv2mesh(output_levels[i][0]).numpy() for i in range(len(output_levels))]
mesh_orig_posed = batch['mesh_posed'].numpy()
body = batch['mesh_body_posed'].numpy()
mesh_pred_levels = [uv2mesh(output_levels[i][0]).detach().cpu().numpy() for i in range(len(output_levels))]
mesh_orig_posed = batch['mesh_posed'].cpu().numpy()
body = batch['mesh_body_posed'].cpu().numpy()
orig_img = (uv_orig + 1) / 2
uv_static_img = (batch['uv_static'].numpy() + 1) / 2
uv_body_posed_img = (batch['uv_body_posed'].numpy() + 1) / 2
uv_curvature = (np.tile(batch['uv_curvature'].numpy().argmax(-1)[:,:,:,np.newaxis], [1,1,1,3]) / 6)
pred_img = np.concatenate([tf.image.resize(output_levels[i][0], (512, 256), tf.image.ResizeMethod.LANCZOS5) for i in range(len(output_levels))], axis=2)
pred_img = (pred_img - pred_img.min(axis=(1,2,3))[:,np.newaxis,np.newaxis,np.newaxis]) / (pred_img.max(axis=(1,2,3))[:,np.newaxis,np.newaxis,np.newaxis] - pred_img.min(axis=(1,2,3))[:,np.newaxis,np.newaxis,np.newaxis])
img = tf.concat((uv_static_img, uv_body_posed_img, uv_curvature, orig_img, pred_img), axis=2)
imgs = [wandb.Image(img[i]) for i in range(4)]
orig_img = (torch.permute(uv_orig, [0,2,3,1]) + 1) / 2
uv_static_img = (torch.permute(batch['uv_static'], [0,2,3,1]) + 1) / 2
uv_body_posed_img = (torch.permute(batch['uv_body_posed'], [0,2,3,1]) + 1) / 2
uv_curvature = (torch.tile(torch.argmax(torch.permute(batch['uv_curvature'], [0,2,3,1]), dim=-1)[:,:,:,None], [1,1,1,3]) / 6)
pred_img = torch.concatenate([torch.permute(F.interpolate(output_levels[i][0], size=(512, 256), mode='bilinear'), [0,2,3,1]) for i in range(len(output_levels))], dim=3)
# pred_img = (pred_img - pred_img.min(axis=(1,2,3))[:,np.newaxis,np.newaxis,np.newaxis]) / (pred_img.max(axis=(1,2,3))[:,np.newaxis,np.newaxis,np.newaxis] - pred_img.min(axis=(1,2,3))[:,np.newaxis,np.newaxis,np.newaxis])
pred_img = (pred_img - pred_img.flatten().reshape(pred_img.shape[0], -1).min(1)[0][:,None,None,None]) / (pred_img.flatten().reshape(pred_img.shape[0], -1).max(1)[0][:,None,None,None] - pred_img.flatten().reshape(pred_img.shape[0], -1).min(1)[0][:,None,None,None])
F = np.load("custom_utils/smpl_faces" + skirt_modifier + ("_orig" if config['is_cape'] else "") + ".npy")
mask = batch['mesh_mask'].numpy().astype(bool)
_, faces = maskFacesBatch(np.ones_like(mesh_orig), F, mask)
img = torch.concat((uv_static_img, uv_body_posed_img, uv_curvature, orig_img, pred_img), axis=2)
imgs = [wandb.Image(img[i].cpu().numpy()) for i in range(4)]
_F = np.load("smpl/smpl_faces" + skirt_modifier + ("_orig" if config['is_cape'] else "") + ".npy")
mask = batch['mesh_mask'].cpu().numpy().astype(bool)
_, faces = maskFacesBatch(np.ones_like(mesh_orig), _F, mask)
if config['use_skinning']:
shape = batch['pose_shape'][:, 72:]
gender = tf.cast(tf.greater_equal(shape[:, 10] , 0.5), tf.int32)
shape_offsets = tf.stack([tf.gather(tf.tensordot(shape[i,:10][tf.newaxis, :], smpl.shapedirs[gender[i]], axes=[[1], [2]])[0,:,:], smpl.to_keep, axis=0) for i in range(shape.shape[0])]).numpy()
gender = torch.greater_equal(shape[:, 10] , 0.5).to(torch.long)
shape_offsets = torch.stack([torch.tensordot(shape[i,:10][None, :], smpl.shapedirs[gender[i]], dims=[[1], [2]])[0,smpl.to_keep,:] for i in range(shape.shape[0])]).cpu().numpy()
gt = [mesh_orig[i][mask[i], :] + shape_offsets[i, mask[i], :] for i in range(mask.shape[0])]
mesh_unposed_shaped_levels = [[] for _ in range(mask.shape[0])]
for i in range(len(output_levels)):
mesh_unposed_shaped = [mesh_pred_levels[i][j, mask[j], :] + shape_offsets[j, mask[j]] for j in range(mask.shape[0])]
for k in range(mask.shape[0]):
mesh_unposed_shaped_levels[k].append(mesh_unposed_shaped[k])
# full res version
mesh_unposed_shaped_full = np.zeros((mask.shape[0], config['smpl_res'], 3), dtype=np.float32)
for i in range(mask.shape[0]):
mesh_unposed_shaped_full[i, mask[i], :] = mesh_unposed_shaped[i]
mesh_posed = smpl.run_garment(batch['pose_shape'][:, :72], batch['pose_shape'][:, 72:], mesh_unposed_shaped_full).numpy()
mesh_posed = smpl(batch['pose_shape'][:, :72], batch['pose_shape'][:, 72:], torch.tensor(mesh_unposed_shaped_full, device=config['device']))[0].cpu().numpy()
else:
mesh_posed = mesh_pred_levels[-1]
gt = [mesh_orig[i][mask[i], :] for i in range(mesh_posed.shape[0])]
@@ -467,7 +484,7 @@ def plot_everything(batch: dict, type: str):
gt_posed = [mesh_orig_posed[i, mask[i], :] for i in range(mask.shape[0])]
mesh_posed_fixed = fix_collision(mesh_posed, body, mask).numpy()
mesh_posed_fixed = fix_collision(torch.Tensor(mesh_posed).to('cuda'), torch.Tensor(body).to('cuda'), torch.Tensor(mask).to('cuda')).cpu().numpy()
mesh_posed = [mesh_posed[j, mask[j], :] for j in range(mask.shape[0])]
@@ -477,7 +494,7 @@ def plot_everything(batch: dict, type: str):
collision_color = [np.ones_like(mesh_posed[i], dtype=np.int32)*230 for i in range(len(mesh_posed))]
else:
collision = [np.concatenate((body[i], mesh_posed[i]), axis=0) for i in range(len(mesh_posed))]
collision_faces = [np.concatenate((F, faces[i] + config['smpl_res']), axis=0) for i in range(len(mesh_posed))]
collision_faces = [np.concatenate((_F, faces[i] + config['smpl_res']), axis=0) for i in range(len(mesh_posed))]
collision_color = [np.concatenate((np.ones_like(body[i], dtype=np.int32)*100, np.ones_like(mesh_posed[i], dtype=np.int32)*230), axis=0) for i in range(len(mesh_posed))]
figs = { type + "/garment_" + str(i): plot_3d(gt[i], mesh_unposed_shaped_levels[i], faces[i], gt_posed[i], collision[i], collision_faces[i], collision_color[i]) for i in range(4) }
@@ -487,7 +504,7 @@ def plot_everything(batch: dict, type: str):
else:
garment_posed_fixed = [mesh_posed_fixed[i, mask[i], :] for i in range(body.shape[0])]
collision_fixed = [np.concatenate((body[i], garment_posed_fixed[i]), axis=0) for i in range(len(mesh_posed))]
gt_posed = [batch['mesh_posed'].numpy()[i, mask[i], :] for i in range(mask.shape[0])]
gt_posed = [batch['mesh_posed'].cpu().numpy()[i, mask[i], :] for i in range(mask.shape[0])]
collision_gt = [np.concatenate((body[i], gt_posed[i]), axis=0) for i in range(len(gt_posed))]
figs_coll = { type + "/garment_" + str(i) + '_collision': plot_3d_collision(collision_gt[i], collision[i], collision_fixed[i], collision_faces[i], collision_color[i]) for i in range(4) }
@@ -505,7 +522,7 @@ def sample(folder):
pred = model.sample(batch['uv_static'][0], batch['uv_body_posed'][0], uv_curvature[0], batch['uv_mask'][0], nr_of_samples=nr_of_samples)
pose, shape = batch['pose_shape'][0, :72], batch['pose_shape'][0, 72:]
pose, shape = tf.tile(pose[tf.newaxis,:], [nr_of_samples,1]), tf.tile(shape[tf.newaxis,:], [nr_of_samples,1])
pose, shape = tf.tile(pose[None,:], [nr_of_samples,1]), tf.tile(shape[None,:], [nr_of_samples,1])
mesh_pred = uv2mesh(tf.transpose(pred, [0,3,1,2]))
mesh_posed = smpl.run_garment(pose, shape, mesh_pred).numpy()
@@ -706,13 +723,13 @@ def generate_results():
shape = batch['pose_shape'][:, 72:]
gender = tf.cast(tf.greater_equal(shape[:, 10] , 0.5), tf.int32)
shape_offsets = tf.stack([tf.gather(tf.tensordot(shape[i,:10][tf.newaxis, :], smpl.shapedirs[gender[i]], axes=[[1], [2]])[0,:,:], smpl.to_keep, axis=0) for i in range(shape.shape[0])])
shape_offsets = tf.stack([tf.gather(tf.tensordot(shape[i,:10][None, :], smpl.shapedirs[gender[i]], axes=[[1], [2]])[0,:,:], smpl.to_keep, axis=0) for i in range(shape.shape[0])])
i = -1
pred = out_levels[i][0]
mesh_pred = uv2mesh(pred)
mesh_pred_shaped = mesh_pred + shape_offsets * batch['mesh_mask'][:,:,tf.newaxis]
mesh_pred_shaped = mesh_pred + shape_offsets * batch['mesh_mask'][:,:,None]
mesh_pred_posed = smpl.run_garment(batch['pose_shape'][:, :72], batch['pose_shape'][:, 72:], mesh_pred_shaped)
for i in range(batch['id'].shape[0]):
@@ -725,157 +742,117 @@ def generate_results():
pass
@tf.function
def speed_step(levels_input, conds):
return model.sample([levels_input, conds], training=False)
def train():
global g_balancing_coeff
global step
valid_iterator = iter(valid_dataloader)
for batch in train_dataloader:
train_loss = train_step(batch)
train_accumulator.update(train_loss)
if step % 25 == 0:
data_valid = next(valid_iterator)
valid_loss = valid_step(data_valid)
valid_accumulator.update(valid_loss)
if step % 1000 == 0 or (step < 1000 and step % 100 == 0):
to_log = dict({ **train_loss, **valid_loss, **{ 'current_step': step } })
if step % 5000 == 0 or (step < 5000 and step % 1000 == 0):
train_figs = plot_everything(batch, 'train')
valid_figs = plot_everything(data_valid, 'valid')
to_log = { **to_log, **train_figs, **valid_figs }
if step == 0:
to_log = { **to_log, **{ 'model/summary': wandb.Html(get_summary_html(model)) } }
if not args['debug']:
wandb.log(to_log, step=step)
train_accumulator.reset()
valid_accumulator.reset()
print(step)
if train_loss['train/euclid_error'] < config['balancing_activation_limit']:
g_balancing_coeff += 1 / 10000
if step % 10000 == 0:
ckpt_saver(model, step, optimizer, None)
step += 1
if __name__ == '__main__':
# init_wandb()
parser = argparse.ArgumentParser(description='')
parser.add_argument('--name', type=str, default='lhunor')
args = vars(parser.parse_args())
if not args['debug']:
init_wandb()
# config_file_names = [key for key in sys.modules.keys() if key.startswith('configs.')]
# assert len(config_file_names) == 1
# config_file_name = config_file_names[0]
# # assert config_file_name.split('.')[1].endswith(args.name)
# assert config_file_name.split('.')[1].endswith(config['name'])
smpl = SMPL(orig_smpl=config['is_cape'])
use_amp = True
smpl = SMPL(orig_smpl=config['is_cape'], path='smpl', device=config['device'])
skirt_modifier = '_skirt' if config['is_skirt'] else ''
dataset_resolutions = config['pyramid']['resolutions'].copy()
if 512 not in dataset_resolutions:
dataset_resolutions.append(512)
factory = DatasetFactory()
dataset_train = factory.get_dataset(folder=config['train_dir'],
batch_size=config['batch_size'],
# batch_size=1,
signature=config['signature_dynamic'],
resolutions=dataset_resolutions,
in_memory=config['dataset_in_memory'],
is_cape=config['is_cape'],
frames_file=config['frames_file'] if 'frames_file' in config.keys() else None)
static_dataset_train = StaticDataset(folder=config['train_dir'],
signature=config['signature_static'],
resolutions=dataset_resolutions,
in_memory=True,
is_cape=config['is_cape'])
dataset_valid = factory.get_dataset(folder=config['valid_dir'],
batch_size=config['batch_size'],
signature=config['signature_dynamic'],
resolutions=dataset_resolutions,
in_memory=config['dataset_in_memory'],
is_cape=config['is_cape'],
frames_file=config['frames_file_valid'] if 'frames_file_valid' in config.keys() else None)
static_dataset_valid = StaticDataset(folder=config['valid_dir'],
signature=config['signature_static'],
resolutions=dataset_resolutions,
in_memory=True,
is_cape=config['is_cape'])
dataset_train = Dataset(config['train_dir'], config['signature'], device=config['device'], frames_file=config['train_frames'], use_cache=config['dataset_in_memory'], resolutions=dataset_resolutions, is_cape=config['is_cape'])
dataset_valid = Dataset(config['valid_dir'], config['signature'], device=config['device'], frames_file=config['valid_frames'], use_cache=config['dataset_in_memory'], resolutions=dataset_resolutions, is_cape=config['is_cape'])
train_dataloader = DataLoader(dataset_train, batch_size=4, drop_last=True, shuffle=True, )
valid_dataloader = DataLoader(dataset_valid, batch_size=4, drop_last=True, shuffle=True)
train_accumulator = Accumulator()
valid_accumulator = Accumulator()
model: Pyramid = Pyramid(config)
optimizer = tf.optimizers.Adam(learning_rate=config['lr'])
balancing_coeff = config['starting_balancing_coeff']
optimizer = optim.Adam(model.parameters(), lr=config['lr'])
_MSSSIM = MS_SSIM(data_range=2, size_average=True, channel=3)
ckpt_saver = CheckpointSaver('checkpoints/iccv/torch/' + config['name'], top_n=5)
scaler = torch.cuda.amp.GradScaler(enabled=use_amp)
g_balancing_coeff = config['starting_balancing_coeff']
reg_coeff = 1
valid_iterator = iter(dataset_valid)
gamma = 1
balancing_activated_step = 0
# sample('checkpoints/cvae_13')
# test()
# generate_results()
ckpt = tf.train.Checkpoint(model=model)
manager = tf.train.CheckpointManager(ckpt, 'checkpoints/' + config['name'], max_to_keep=20)
# load_model(model, 'checkpoints/pyr_abl_32_skirt', iteration='180000')
step = 0
if config['pretrained_model'] != '':
# assert len(config['pyramid']['resolutions']) == 1
load_model(model, config['pretrained_model'], iteration=config['pretrained_model_iteration'])
step = load_model(model, config['pretrained_model'], optimizer=optimizer, scaler=scaler, iteration=config['pretrained_model_iteration'])
if not args['debug']:
model = torch.compile(model)
wandb.watch(model, log='all', log_freq=1000, log_graph=True)
if len(config['pyramid']['resolutions']) == 1:
model.static_enc.trainable = False
model.body_enc.trainable = False
model.curv_enc.trainable = False
for param in model.static_enc.parameters(): param.requires_grad = False
for param in model.body_enc.parameters(): param.requires_grad = False
for param in model.curv_enc.parameters(): param.requires_grad = False
else:
load_model(model, config['pyramid']['pretrained_pyramid'], iteration=(config['pyramid']['pretrained_pyramid_iteration'] if 'pretrained_pyramid_iteration' in config.keys() else ''))
model.static_enc.trainable = False
model.body_enc.trainable = False
model.curv_enc.trainable = False
model.cond_encoder.trainable = False
for param in model.static_enc.parameters(): param.requires_grad = False
for param in model.body_enc.parameters(): param.requires_grad = False
for param in model.curv_enc.parameters(): param.requires_grad = False
# load_model(model.levels[0].upscale_enhancer, config['pyramid']['levels'][0]['upscale_enhancer'])
# model.levels[0].upscale_enhancer.trainable = False
# for i in range(len(model.levels)-1):
# load_model(model.levels[i].upscale_enhancer, config['pyramid']['levels'][i]['upscale_enhancer'])
# model.levels[i].upscale_enhancer.trainable = False
# if config['pyramid']['resolutions'][-1] != 512:
# load_model(model.upscale_enhancer, config['pyramid']['final_upscale_enhancer'])
# model.upscale_enhancer.trainable = False
for i in range(len(model.levels)-1):
load_model(model.levels[i].upscale_enhancer, config['pyramid']['levels'][i]['upscale_enhancer'])
model.levels[i].upscale_enhancer.trainable = False
if config['pyramid']['resolutions'][-1] != 512:
load_model(model.upscale_enhancer, config['pyramid']['final_upscale_enhancer'])
model.upscale_enhancer.trainable = False
for i in range(len(model.levels)-1):
model.levels[i].trainable = False
for param in model.levels[i].parameters(): param.requires_grad = False
# load_model(model.cond_encoder, config['pyramid']['pretrained_pyramid'])
# load_model(model.levels[0], config['pyramid']['pretrained_pyramid'])
# load_model(model.levels[-1], 'checkpoints/cvae_19')
if not args['debug']:
model = torch.compile(model)
wandb.watch(model, log='all', log_freq=1000, log_graph=True)
step = 0
times = []
for batch in dataset_train:
batch = static_dataset_train.expand_batch(batch_to_dict(batch, config['signature_dynamic']))
# levels_input, conds = group_inputs_for_levels(batch, config)
# start = time.perf_counter()
# out = speed_step(levels_input, conds)
# t = time.perf_counter() - start
# print(t)
# times.append(t)
# if step > 200:
# break
train_loss = train_step(batch)
data_valid = valid_iterator.get_next()
data_valid = static_dataset_valid.expand_batch(batch_to_dict(data_valid, config['signature_dynamic']))
valid_loss = valid_step(data_valid)
to_log = { **train_loss, **valid_loss, **{ 'current_step': step} }
if step % 5000 == 0 or (step < 5000 and step % 500 == 0):
train_figs = plot_everything(batch, 'train')
valid_figs = plot_everything(data_valid, 'valid')
to_log = { **to_log, **train_figs, **valid_figs }
if step % 5000 == 0:
manager.save(checkpoint_number=step)
if step == 0:
to_log = { **to_log, **{ 'model/summary': wandb.Html(get_summary_html(model)) } }
if train_loss['train/euclid_error'] < 0.004:
if balancing_activated_step == 0:
balancing_activated_step = step
balancing_coeff = balancing_coeff + 1 / 20000
to_log = { **to_log, **{ 'train/balancing_coeff': balancing_coeff, 'train/gamma': gamma } }
wandb.log(to_log, step = step)
print(step)
step += 1
# print('mean:')
# print(np.mean(times[1:]))
train()