From 83294bc00cc6bc83be6effc2c036fba8eca36a64 Mon Sep 17 00:00:00 2001 From: Hunor Laczko Date: Mon, 26 Sep 2022 12:54:03 +0000 Subject: [PATCH] update cvae_low version --- configs/config_cvae.py | 26 +++++++- cvae.py | 8 ++- cvae_low.py | 148 +++++++---------------------------------- 3 files changed, 53 insertions(+), 129 deletions(-) diff --git a/configs/config_cvae.py b/configs/config_cvae.py index 1c5ecea..e71f6ed 100644 --- a/configs/config_cvae.py +++ b/configs/config_cvae.py @@ -1,7 +1,7 @@ from configs.config_base import config -config['gpu'] = '3' -config['name'] = 'cvae_19' +config['gpu'] = '0' +config['name'] = 'cvae_19_2' config['notes'] = '\ offsets over average unposed garment. remove global orientation (in posed body input and in pose_shape parameters. \ retrained body encoder with deoriented bodies \ @@ -11,9 +11,17 @@ remove shape from unposed garment\ training on 100 garments with offsets over the average of these\ fixed autoencoders!!!!!!!!!\ cvae only loads encoders instead AEs\ +with edge and collision loss\ ' +# config['resolution'] = 32 + +if config['resolution'] != 512: + config['notes'] += '\ +using custom enhancer for upscaling, using not filled versions\ +' + config['lr'] = 0.0001 config['starting_balancing_coeff'] = 0.0001 @@ -31,7 +39,19 @@ config['signature_static'] = { 'uv_static': (512, 256, 3), 'uv_mask': (512, 256,), } -config['avg_garment_path'] = 'custom_data/train_100_no_shape_avg_uv.npy' +if config['resolution'] != 512: + res_modifier = '_' + str(config['resolution']) + config['signature_dynamic']['uv_body_posed' + res_modifier] = (config['resolution'], config['resolution']//2, 3) + config['signature_dynamic']['uv_unposed' + res_modifier] = (config['resolution'], config['resolution']//2, 3) + config['signature_dynamic']['uv_curvature' + res_modifier] = (config['resolution'], config['resolution']//2) + config['signature_static']['uv_static' + res_modifier] = (config['resolution'], config['resolution']//2, 3) + config['signature_static']['uv_mask' + res_modifier] = (config['resolution'], config['resolution']//2) + +if config['resolution'] == 512: + config['avg_garment_path'] = 'custom_data/train_100_no_shape_avg_uv.npy' +else: + config['avg_garment_path'] = '' + config['static_model'] = 'checkpoints/ae_static_7_2_x' config['body_model'] = 'checkpoints/ae_body_9_x' config['curvature_model'] = 'checkpoints/ae_curv_9_x' diff --git a/cvae.py b/cvae.py index 205b10a..001bc8d 100644 --- a/cvae.py +++ b/cvae.py @@ -87,21 +87,23 @@ def train_step(batch): 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) - e_loss = tf.reduce_mean(edge_loss(mesh_pred, mesh_orig, batch['mesh_mask'])) - c_loss = tf.reduce_mean(collision_penalty(mesh_pred_posed, batch['mesh_body_posed'], batch['mesh_mask'])) + e_loss = edge_loss(mesh_pred, mesh_orig, batch['mesh_mask']) + c_loss = collision_penalty(mesh_pred_posed, batch['mesh_body_posed'], batch['mesh_mask']) l1_loss = tf.reduce_sum(tf.reduce_sum((tf.math.abs(batch['uv_unposed'] - pred)), axis=-1) * batch['uv_mask'], axis=[1, 2]) l2_loss = 1e3 * tf.reduce_mean(tf.reduce_sum(((batch['uv_unposed'] - pred) ** 2), axis=-1) * batch['uv_mask'], axis=[1, 2]) # ms_ssim = tf.image.ssim_multiscale(tf.transpose(pred, [0,2,3,1]) + 1, tf.transpose(batch['uv_unposed'], [0,2,3,1]) + 1, max_val=2) ms_ssim = tf.image.ssim_multiscale(pred + 1, batch['uv_unposed'] + 1, max_val=3, power_factors=(0.5, 0.2, 0.1, 0.1, 0.1)) # ms_ssim = tf.reduce_mean(tf.image.ssim_multiscale((tf.transpose(pred, [0,2,3,1]) + 1) * batch['uv_mask'][:,:,:,tf.newaxis], (tf.transpose(batch['uv_unposed'], [0,2,3,1]) + 1) * batch['uv_mask'][:,:,:,tf.newaxis], max_val=2)) - reconstr_loss = l2_loss + reconstr_loss = l2_loss + e_loss + c_loss reconstr_loss = tf.reduce_mean(reconstr_loss) kl = kl_loss(mean, logvar) loss = reconstr_loss + balancing_coeff * kl ms_ssim = tf.reduce_mean(ms_ssim) l1_loss = tf.reduce_mean(l1_loss) l2_loss = tf.reduce_mean(l2_loss) + e_loss = tf.reduce_mean(e_loss) + c_loss = tf.reduce_mean(c_loss) mean = tf.reduce_mean(mean) logvar = tf.reduce_mean(logvar) diff --git a/cvae_low.py b/cvae_low.py index d8ef11f..85e6405 100644 --- a/cvae_low.py +++ b/cvae_low.py @@ -1,5 +1,6 @@ import os -os.environ["CUDA_VISIBLE_DEVICES"] = "1" +from configs.config_cvae import config +os.environ["CUDA_VISIBLE_DEVICES"] = config['gpu'] import tensorflow as tf gpus = tf.config.experimental.list_physical_devices('GPU') @@ -29,86 +30,29 @@ from tf_utils import check_tf_version, allow_memory_growth, split_gpu_for_testin from losses import d_logistic, d_logistic_r1_reg, g_logistic_non_saturating, g_logistic_ns_pathreg, edge_loss, uv_mask_loss, bend_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 custom_utils.custom_utils import batch_to_dict, load_model, plot_3d from dataset import DatasetFactory, StaticDataset from custom_utils.smpl import SMPL -from models.cvae_model import CVAE#, CVAE128, CVAE32 +from models.cvae_model import CVAE from models.upscaler import UpscalerBasic as Upscaler -name='cvae_low_4_2_d32' def init_wandb(): - run = wandb.init(project='vae', name=name, - notes='learn offsets. smaller flattened latent vectors. with 25-85k pretraining. smaller batchsize. fixed balancing of reconst and kl loss for baseline. \ -new implementation for kl loss. no lr schedule. 25 jumpsuits, average reconst_loss before calculation, linearly increasing balancing, up to 1 from 0.1. \ -offsets over average unposed garment. remove global orientation (in posed body input and in pose_shape parameters. \ -retrained body encoder with deoriented bodies \ -use mean l2 loss. small batch. l2 * 1000, balancing to 1 from 0.01\ -fixed gender issue and retrained body encoder\ -remove shape from unposed garment\ -training on 100 garments , absolute, not offsets\ -using custom enhancer for upscaling, using not filled versions\ -calculating loss at full res') + run = wandb.init(project='vae', name=config['name'], notes=config['notes']) artifact = wandb.Artifact('code', type='code') - artifact.add_file('cvae.py') + artifact.add_file('cvae_low.py') artifact.add_file('models/cvae_model.py') - artifact.add_file('dataset.py') artifact.add_file('losses.py') + artifact.add_file('configs/config_base.py') + artifact.add_file('configs/config_cvae.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') run.log_artifact(artifact) uv2mesh = UV2Mesh(res=256) -def plot_3d(mesh_gt, mesh_pred_unposed, faces, collision, collision_faces, collision_color): - mesh_pred_unposed[:,2] += 1 - collision[:,2] += 2 - fig = go.Figure(data=[ - go.Mesh3d( - # 8 vertices of a cube - x=mesh_gt[:,0], - y=mesh_gt[:,2], - z=mesh_gt[:,1], - # i, j and k give the vertices of triangles - i = faces[:,0], - j = faces[:,1], - k = faces[:,2] - ), - go.Mesh3d( - # 8 vertices of a cube - x=mesh_pred_unposed[:,0], - y=mesh_pred_unposed[:,2], - z=mesh_pred_unposed[:,1], - # i, j and k give the vertices of triangles - i = faces[:,0], - j = faces[:,1], - k = faces[:,2], - ), - go.Mesh3d( - # 8 vertices of a cube - x=collision[:,0], - y=collision[:,2], - z=collision[:,1], - # i, j and k give the vertices of triangles - i = collision_faces[:,0], - j = collision_faces[:,1], - k = collision_faces[:,2], - vertexcolor=collision_color - ) - ]) - - fig.update_layout(scene_aspectmode='manual', - scene_aspectratio=dict(x=1, y=2, z=1)) - fig.update_layout( - scene = dict( - xaxis = dict(nticks=4, range=[-1,1],), - yaxis = dict(nticks=4, range=[-1,3],), - zaxis = dict(nticks=4, range=[-1,1],),) - ) - - return fig - - - def log_normal_pdf(sample, mean, logvar, raxis=1): log2pi = tf.math.log(2. * np.pi) @@ -131,11 +75,10 @@ def train_step(batch): uv_curvature_low = tf.one_hot(batch['uv_curvature' + res_modifier], 7) pred, mean, logvar, z = model([batch['uv_unposed' + res_modifier], batch['uv_static' + res_modifier], batch['uv_body_posed' + res_modifier], uv_curvature_low, batch['uv_mask' + res_modifier], batch['uv_static'], batch['uv_body_posed'], uv_curvature, batch['uv_mask']], use_mask=True) - # pred = pred * batch['uv_mask' + res_modifier][:,:,:,tf.newaxis] pred_enhanced = enhance(upscale_enhancer, pred, batch['uv_mask' + res_modifier], training=False) pred_full_res = tf.image.resize(pred_enhanced, size=(512, 256), method=tf.image.ResizeMethod.LANCZOS5) - mesh_pred = uv2mesh(pred_enhanced) + mesh_pred = uv2mesh(pred_full_res) mesh_orig = batch['mesh_unposed'] shape = batch['pose_shape'][:, 72:] @@ -153,7 +96,7 @@ def train_step(batch): ms_ssim = tf.image.ssim(pred + 1, batch['uv_unposed' + res_modifier] + 1, max_val=3) ms_ssim_full = tf.image.ssim(pred_full_res + 1, batch['uv_unposed'] + 1, max_val=3) else: - ms_ssim = tf.image.ssim_multiscale(pred + 1, batch['uv_unposed' + res_modifier] + 1, max_val=3, power_factors=(0.5, 0.2, 0.1, 0.1, 0.1)) + ms_ssim = tf.image.ssim_multiscale(pred + 1, batch['uv_unposed' + res_modifier] + 1, max_val=3, power_factors=(0.5, 0.3, 0.1, 0.1)) reconstr_loss = l2_loss_full reconstr_loss = tf.reduce_mean(reconstr_loss) @@ -167,18 +110,12 @@ def train_step(batch): mean = tf.reduce_mean(mean) logvar = tf.reduce_mean(logvar) - # mesh_orig_2 = uv2mesh(batch['uv_unposed' + res_modifier]) 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']) - # metric_2 = tf.reduce_sum(tf.math.sqrt(tf.reduce_sum(batch['mesh_mask'][:,:,tf.newaxis] * (mesh_orig_2 - mesh_pred) ** 2, axis=-1))) / tf.reduce_sum(batch['mesh_mask']) - euclid_posed = 0 # tf.reduce_sum(tf.math.sqrt(tf.reduce_sum(batch['mesh_mask'][:,:,tf.newaxis] * (mesh_skinned - batch['mesh_posed']) ** 2, axis=-1))) / tf.reduce_sum(batch['mesh_mask']) gradients = tape.gradient(loss, model.trainable_variables) - # gradients = [tf.clip_by_norm(g, 1.0) for g in gradients] - # gradients = [tf.clip_by_value(g, -1.0, 1.0) for g in gradients] optimizer.apply_gradients(zip(gradients, model.trainable_variables)) return { 'train/loss': loss, 'train/edge_loss': e_loss, 'train/collision_loss': c_loss, 'train/l1_loss': l1_loss, 'train/euclid_error': metric, 'train/ms_ssim': ms_ssim, - #'train/euclid_error_reconst': metric_2, 'train/l2_loss_full_res': l2_loss_full, 'train/ms_ssim_full_res': ms_ssim_full, - 'train/euclid_error_posed': euclid_posed, 'train/reconstr_loss': reconstr_loss, + 'train/reconstr_loss': reconstr_loss, 'train/mean': mean, 'train/logvar': logvar, 'train/kl_loss': kl, 'train/var': np.e ** logvar, 'train/reg_coeff': reg_coeff, 'train/l2_loss': l2_loss } @@ -226,8 +163,8 @@ def valid_step(batch): def plot_everything(batch: dict, type: str): uv_curvature = tf.one_hot(batch['uv_curvature'], 7) uv_curvature_low = tf.one_hot(batch['uv_curvature' + res_modifier], 7) - pred, mean, logvar, z = model([batch['uv_unposed' + res_modifier], batch['uv_static' + res_modifier], batch['uv_body_posed' + res_modifier], uv_curvature_low, batch['uv_static'], batch['uv_body_posed'], uv_curvature]) - # pred = pred * batch['uv_mask' + res_modifier][:,:,:,tf.newaxis] + pred, mean, logvar, z = model([batch['uv_unposed' + res_modifier], batch['uv_static' + res_modifier], batch['uv_body_posed' + res_modifier], uv_curvature_low, + batch['uv_mask' + res_modifier], batch['uv_static'], batch['uv_body_posed'], uv_curvature, batch['uv_mask']], use_mask=True) pred = enhance(upscale_enhancer, pred, batch['uv_mask' + res_modifier], training=False) mesh_pred = uv2mesh(pred).numpy() @@ -433,64 +370,31 @@ if __name__ == '__main__': res_modifier = '_' + str(resolution) factory = DatasetFactory() - signature_dynamic = { 'id': (), - 'pose_shape': (72 + 11,), - # 'uv_posed': (512, 256, 3), - 'uv_body_posed': (512, 256, 3), - 'uv_body_posed' + res_modifier: (resolution, resolution//2, 3), - 'mesh_body_posed': (14475, 3,), - 'mesh_posed': (14475, 3,), - 'mesh_unposed': (14475, 3,), - 'uv_unposed': (512, 256, 3), - 'uv_unposed' + res_modifier: (resolution, resolution//2, 3), - 'uv_curvature': (512, 256,), - 'uv_curvature' + res_modifier: (resolution, resolution//2), - } - signature_static = { 'uv_static': (512, 256, 3), - 'uv_static' + res_modifier: (resolution, resolution//2, 3), - # 'mesh_static': (14475, 3,), - 'mesh_mask': (14475,), - 'uv_mask' + res_modifier: (resolution, resolution//2,), - 'uv_mask': (512, 256,), - } - dataset_train = factory.get_dataset(folder='/home/workspace/local-data/train_100', batch_size=batch_size, signature=signature_dynamic, use_filled=False, resolutions=[512, resolution]) - static_dataset_train = StaticDataset(folder='/home/workspace/local-data/train_100', signature=signature_static, use_filled=False, resolutions=[512, resolution]) - dataset_valid = factory.get_dataset(folder='/home/workspace/local-data/jumpsuit_20/valid', batch_size=batch_size, signature=signature_dynamic) - static_dataset_valid = StaticDataset(folder='/home/workspace/local-data/jumpsuit_20/valid', signature=signature_static) + dataset_train = factory.get_dataset(folder=config['train_dir'], batch_size=config['batch_size'], signature=config['signature_dynamic'], use_filled=config['use_filled'], resolutions=[512, config['resolution']]) + static_dataset_train = StaticDataset(folder=config['train_dir'], signature=config['signature_static'], use_filled=config['use_filled'], resolutions=[512, config['resolution']]) + dataset_valid = factory.get_dataset(folder=config['valid_dir'], batch_size=config['batch_size'], signature=config['signature_dynamic']) + static_dataset_valid = StaticDataset(folder=config['valid_dir'], signature=config['signature_static']) - - model = CVAE(#avg_garment_path='custom_data/train_100_no_shape_avg_uv.npy', - size=resolution, - latent_dim=128, static_model='checkpoints/ae_static_6_copy', - body_model='checkpoints/ae_body_8_copy', - curvature_model='checkpoints/ae_curvature_7_copy', - freeze_encoders=True, - encoder_res=512) + model = CVAE(config) upscale_enhancer = get_upscale_enhancer('checkpoints/up_2_3', res=resolution) - - # lr_schedule = tf.keras.optimizers.schedules.ExponentialDecay(0.0001, decay_steps=30000, decay_rate=0.1, staircase=False) - optimizer = tf.optimizers.Adam(learning_rate=0.0001) - balancing_coeff = 0.0001 + optimizer = tf.optimizers.Adam(learning_rate=config['lr']) + balancing_coeff = config['starting_balancing_coeff'] reg_coeff = 1 - folder = 'models/ae_logs/' valid_iterator = iter(dataset_valid) gamma = 1 balancing_activated_step = 0 - # avg_jumpsuit = np.load('custom_data/jumpsuit_20_avg_uv.npy') - # sample('checkpoints/cvae_13') - ckpt = tf.train.Checkpoint(model=model) - manager = tf.train.CheckpointManager(ckpt, 'checkpoints/' + name, max_to_keep=3) + manager = tf.train.CheckpointManager(ckpt, 'checkpoints/' + config['name'], max_to_keep=3) step = 0 for batch in dataset_train: - batch = static_dataset_train.expand_batch(batch_to_dict(batch, signature_dynamic)) + batch = static_dataset_train.expand_batch(batch_to_dict(batch, config['signature_dynamic'])) train_loss = train_step(batch) @@ -514,8 +418,6 @@ if __name__ == '__main__': balancing_activated_step = step balancing_coeff = balancing_coeff + 1 / 15000 - # gamma = 0.99 * gamma + 0.01 * train_loss['train/reconstr_loss'] - to_log = { **to_log, **{ 'train/balancing_coeff': balancing_coeff, 'train/gamma': gamma } } wandb.log(to_log, step = step)