emjay73
commited on
Commit
·
fafe050
1
Parent(s):
3bb64b4
add checkpoints
Browse files- tcan/first_stage/models/[AppearanceEncoder]_[Epoch=1250]_[Iter=17500]_[loss=0.0864].ckpt +3 -0
- tcan/first_stage/models/[UNet]_[Epoch=1250]_[Iter=17500]_[loss=0.0864].ckpt +3 -0
- tcan/second_stage/args.json +50 -0
- tcan/second_stage/config.yaml +91 -0
- tcan/second_stage/models/[ControlNetT]_[Epoch=74]_[Iter=12500]_[loss=0.0556].ckpt +3 -0
- tcan/second_stage/models/[UNet]_[Epoch=74]_[Iter=12500]_[loss=0.0556].ckpt +3 -0
tcan/first_stage/models/[AppearanceEncoder]_[Epoch=1250]_[Iter=17500]_[loss=0.0864].ckpt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:bc3a283b5a5cff45492462c2e778efe47411444d08670ffd5a4ca98e3f72ef11
|
| 3 |
+
size 897318912
|
tcan/first_stage/models/[UNet]_[Epoch=1250]_[Iter=17500]_[loss=0.0864].ckpt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:3437ecffddc8ffb74cf5aeb2652d37cc1416adf19cba6b95d37d64bf792e7031
|
| 3 |
+
size 897318912
|
tcan/second_stage/args.json
ADDED
|
@@ -0,0 +1,50 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"config": "./configs/train/reimple_second_stage_v2.yaml",
|
| 3 |
+
"batch_size": 1,
|
| 4 |
+
"motion_type": "dwpose",
|
| 5 |
+
"is_second_stage": true,
|
| 6 |
+
"freeze_controlnet": false,
|
| 7 |
+
"use_HAP": false,
|
| 8 |
+
"use_augchibi": false,
|
| 9 |
+
"ref_aug": null,
|
| 10 |
+
"motion_aug": null,
|
| 11 |
+
"image_dataset_name": "TikTokImageDataset2",
|
| 12 |
+
"video_dataset_name": "TikTokVideoDataset",
|
| 13 |
+
"pretrained_vae_path": null,
|
| 14 |
+
"pretrained_appearance_encoder_path": "./logs/from39_20240224_1st_freezectrl_temporalctrl/models/[AppearanceEncoder]_[Epoch=1250]_[Iter=17500]_[loss=0.0864].ckpt",
|
| 15 |
+
"pretrained_controlnet_path": "./pretrained_models/ControlNet-v1-1/control_v11p_sd15_openpose_RenamedForMA.pth",
|
| 16 |
+
"pretrained_model_path": null,
|
| 17 |
+
"pretrained_unet_path": "./logs/from39_20240224_1st_freezectrl_temporalctrl/models/[UNet]_[Epoch=1250]_[Iter=17500]_[loss=0.0864].ckpt",
|
| 18 |
+
"init_unet_lora": true,
|
| 19 |
+
"init_appearance_encoder_lora": false,
|
| 20 |
+
"load_unet_lora_weight": true,
|
| 21 |
+
"load_ae_lora_weight": false,
|
| 22 |
+
"use_learnable_taumap": false,
|
| 23 |
+
"unet_lora_midup": false,
|
| 24 |
+
"save_name": "SecondUnetloraTctrlv2_fromEpoch1250from39",
|
| 25 |
+
"n_epochs": 90,
|
| 26 |
+
"n_iters": 100000,
|
| 27 |
+
"size": null,
|
| 28 |
+
"data_root_dir": "./DATA/TikTok",
|
| 29 |
+
"HAP_root_dir": "./DATA/HAP/train",
|
| 30 |
+
"augchibi_root_dir": "./DATA/sd_augchibi",
|
| 31 |
+
"save_root_dir": "./logs",
|
| 32 |
+
"save_model_iter_freq": 2500,
|
| 33 |
+
"validation_step": 2500,
|
| 34 |
+
"validation_step_lst": [
|
| 35 |
+
3
|
| 36 |
+
],
|
| 37 |
+
"seed": 42,
|
| 38 |
+
"num_workers": 2,
|
| 39 |
+
"use_temporal_controlnet": true,
|
| 40 |
+
"save_dir": "./logs/20240405_SecondUnetloraTctrlv2_fromEpoch1250from39",
|
| 41 |
+
"sample_save_dir": "./logs/20240405_SecondUnetloraTctrlv2_fromEpoch1250from39/samples",
|
| 42 |
+
"model_save_dir": "./logs/20240405_SecondUnetloraTctrlv2_fromEpoch1250from39/models",
|
| 43 |
+
"tb_save_dir": "./logs/20240405_SecondUnetloraTctrlv2_fromEpoch1250from39/tb",
|
| 44 |
+
"args_save_path": "./logs/20240405_SecondUnetloraTctrlv2_fromEpoch1250from39/args.json",
|
| 45 |
+
"config_save_path": "./logs/20240405_SecondUnetloraTctrlv2_fromEpoch1250from39/config.yaml",
|
| 46 |
+
"n_gpus": 2,
|
| 47 |
+
"DDP": true,
|
| 48 |
+
"num_processes": 2,
|
| 49 |
+
"local_rank": 0
|
| 50 |
+
}
|
tcan/second_stage/config.yaml
ADDED
|
@@ -0,0 +1,91 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
appearance_encoder_config_path: pretrained_models/MagicAnimate/appearance_encoder/config.json
|
| 2 |
+
controlnet_config_path: pretrained_models/MagicAnimate/densepose_controlnet/config.json
|
| 3 |
+
pretrained_model_path: pretrained_models/stable-diffusion-v1-5
|
| 4 |
+
pretrained_vae_path: pretrained_models/sd-vae-ft-mse
|
| 5 |
+
pretrained_controlnet_path: ./pretrained_models/ControlNet-v1-1/control_v11p_sd15_openpose_RenamedForMA.pth
|
| 6 |
+
pretrained_appearance_encoder_path: ./logs/from39_20240224_1st_freezectrl_temporalctrl/models/[AppearanceEncoder]_[Epoch=1250]_[Iter=17500]_[loss=0.0864].ckpt
|
| 7 |
+
motion_module: pretrained_models/AnimateDiff/mm_sd_v15_v2.ckpt
|
| 8 |
+
fusion_blocks: midup
|
| 9 |
+
unet_additional_kwargs:
|
| 10 |
+
unet_use_cross_frame_attention: false
|
| 11 |
+
unet_use_temporal_attention: false
|
| 12 |
+
use_motion_module: true
|
| 13 |
+
motion_module_resolutions:
|
| 14 |
+
- 1
|
| 15 |
+
- 2
|
| 16 |
+
- 4
|
| 17 |
+
- 8
|
| 18 |
+
motion_module_mid_block: false
|
| 19 |
+
motion_module_decoder_only: false
|
| 20 |
+
motion_module_type: Vanilla
|
| 21 |
+
motion_module_kwargs:
|
| 22 |
+
num_attention_heads: 8
|
| 23 |
+
num_transformer_block: 1
|
| 24 |
+
attention_block_types:
|
| 25 |
+
- Temporal_Self
|
| 26 |
+
- Temporal_Self
|
| 27 |
+
temporal_position_encoding: true
|
| 28 |
+
temporal_position_encoding_max_len: 32
|
| 29 |
+
temporal_attention_dim_div: 1
|
| 30 |
+
use_learnable_taumap: false
|
| 31 |
+
noise_scheduler_kwargs:
|
| 32 |
+
num_train_timesteps: 1000
|
| 33 |
+
beta_start: 0.00085
|
| 34 |
+
beta_end: 0.012
|
| 35 |
+
beta_schedule: linear
|
| 36 |
+
steps_offset: 1
|
| 37 |
+
clip_sample: false
|
| 38 |
+
learning_rate: 1.0e-05
|
| 39 |
+
adam_beta1: 0.9
|
| 40 |
+
adam_beta2: 0.999
|
| 41 |
+
adam_weight_decay: 0.01
|
| 42 |
+
adam_epsilon: 1.0e-08
|
| 43 |
+
lr_scheduler: constant
|
| 44 |
+
max_grad_norm: 1.0
|
| 45 |
+
seed:
|
| 46 |
+
- 1
|
| 47 |
+
steps: 25
|
| 48 |
+
guidance_scale: 7.5
|
| 49 |
+
source_image:
|
| 50 |
+
- DATA/TikTok/valid_video/00337/images.mp4
|
| 51 |
+
- DATA/TikTok/valid_video/00338/images.mp4
|
| 52 |
+
- DATA/TikTok/valid_video/201_002_1x1/images.mp4
|
| 53 |
+
- DATA/TikTok/valid_video/201_005_1x1/images.mp4
|
| 54 |
+
- DATA/TikTok/valid_video/201_021_1x1/images.mp4
|
| 55 |
+
- DATA/TikTok/valid_video/201_024_1x1/images.mp4
|
| 56 |
+
- DATA/TikTok/valid_video/202_006_1x1/images.mp4
|
| 57 |
+
- DATA/TikTok/valid_video/202_007_1x1/images.mp4
|
| 58 |
+
- DATA/TikTok/valid_video/202_025_1x1/images.mp4
|
| 59 |
+
- DATA/TikTok/valid_video/203_006_1x1/images.mp4
|
| 60 |
+
source_pose:
|
| 61 |
+
- ''
|
| 62 |
+
- ''
|
| 63 |
+
- ''
|
| 64 |
+
- ''
|
| 65 |
+
- ''
|
| 66 |
+
- ''
|
| 67 |
+
- ''
|
| 68 |
+
- ''
|
| 69 |
+
- ''
|
| 70 |
+
- ''
|
| 71 |
+
video_path:
|
| 72 |
+
- DATA/TikTok/valid_video/00337/dwpose.mp4
|
| 73 |
+
- DATA/TikTok/valid_video/00338/dwpose.mp4
|
| 74 |
+
- DATA/TikTok/valid_video/201_002_1x1/dwpose.mp4
|
| 75 |
+
- DATA/TikTok/valid_video/201_005_1x1/dwpose.mp4
|
| 76 |
+
- DATA/TikTok/valid_video/201_021_1x1/dwpose.mp4
|
| 77 |
+
- DATA/TikTok/valid_video/201_024_1x1/dwpose.mp4
|
| 78 |
+
- DATA/TikTok/valid_video/202_006_1x1/dwpose.mp4
|
| 79 |
+
- DATA/TikTok/valid_video/202_007_1x1/dwpose.mp4
|
| 80 |
+
- DATA/TikTok/valid_video/202_025_1x1/dwpose.mp4
|
| 81 |
+
- DATA/TikTok/valid_video/203_006_1x1/dwpose.mp4
|
| 82 |
+
size: 512
|
| 83 |
+
L: 16
|
| 84 |
+
S: 1
|
| 85 |
+
I: 0
|
| 86 |
+
clip: 0
|
| 87 |
+
offset: 0
|
| 88 |
+
max_length: null
|
| 89 |
+
video_type: condition
|
| 90 |
+
invert_video: false
|
| 91 |
+
save_individual_videos: false
|
tcan/second_stage/models/[ControlNetT]_[Epoch=74]_[Iter=12500]_[loss=0.0556].ckpt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:3fb1dbf048a66452a402ce67d84ac996173034b502a05c1b5691a6dc15e9730c
|
| 3 |
+
size 897843200
|
tcan/second_stage/models/[UNet]_[Epoch=74]_[Iter=12500]_[loss=0.0556].ckpt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:b4bc2fd4ff06f9be47f890c321f2c985c67520e5b255beddd85f2a723dd78f6d
|
| 3 |
+
size 898105344
|