Upload 2 files
Browse files- config.yaml +77 -0
- model_best.pth +3 -0
config.yaml
ADDED
@@ -0,0 +1,77 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
train:
|
2 |
+
epoch: 201
|
3 |
+
batchsize: 8
|
4 |
+
lr: 5e-5
|
5 |
+
lr_gamma: 0.1
|
6 |
+
lr_steps: [30, 40]
|
7 |
+
cos: True # use cosine lr schedule
|
8 |
+
checkpoint_every: 3000
|
9 |
+
|
10 |
+
model:
|
11 |
+
target: modules.mage_model.MAGE
|
12 |
+
params:
|
13 |
+
codebook_size: 512
|
14 |
+
frames_length: 10
|
15 |
+
image_resolution: 16
|
16 |
+
vision_width: 512
|
17 |
+
dropout: 0.2
|
18 |
+
use_cids: False
|
19 |
+
randomness: True
|
20 |
+
auto_beta: True
|
21 |
+
v_kl: 100
|
22 |
+
|
23 |
+
first_stage_config:
|
24 |
+
target: ldm.models.autoencoder.AutoencoderKL
|
25 |
+
params:
|
26 |
+
monitor: val/rec_loss
|
27 |
+
embed_dim: 4
|
28 |
+
ckpt_path: "models/autoencoders/kl_f8_cater/last_caterv2.ckpt"
|
29 |
+
ddconfig:
|
30 |
+
double_z: true
|
31 |
+
z_channels: 4
|
32 |
+
resolution: 128
|
33 |
+
in_channels: 3
|
34 |
+
out_ch: 3
|
35 |
+
ch: 128
|
36 |
+
ch_mult:
|
37 |
+
- 1
|
38 |
+
- 2
|
39 |
+
- 4
|
40 |
+
- 4
|
41 |
+
num_res_blocks: 2
|
42 |
+
attn_resolutions: [ ]
|
43 |
+
dropout: 0.0
|
44 |
+
lossconfig:
|
45 |
+
target: torch.nn.Identity
|
46 |
+
text_encoder_config:
|
47 |
+
target: modules.mage_model.TransformerTextEncoder
|
48 |
+
params:
|
49 |
+
vocab_size: 50
|
50 |
+
context_length: 38
|
51 |
+
transformer_width: 512
|
52 |
+
transformer_layers: 2
|
53 |
+
output_dim: 512
|
54 |
+
padding_idx: 0
|
55 |
+
dropout: 0.1
|
56 |
+
ma_config:
|
57 |
+
target: modules.mage_model.MAEncoder
|
58 |
+
params:
|
59 |
+
layers: 1
|
60 |
+
d_model: 512
|
61 |
+
generate_decoder_config:
|
62 |
+
target: modules.mage_model.FlatAxialDecoder
|
63 |
+
params:
|
64 |
+
in_channels: 512
|
65 |
+
out_channels: 4
|
66 |
+
model_channels: 512
|
67 |
+
frames_length: 10
|
68 |
+
layers: 6
|
69 |
+
|
70 |
+
data:
|
71 |
+
target: dataload.CATER
|
72 |
+
params:
|
73 |
+
dataset: 'caterv2'
|
74 |
+
data_root: '../datasets/CATER-GEN-v2' # ../datasets/CATER-GEN-v2
|
75 |
+
frames_length: 10
|
76 |
+
sample_speed: [3.0, 6.0]
|
77 |
+
randomness: True
|
model_best.pth
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:bb5546980514e22e61159bb799a6a05634b84e9d6fb076fdb53e16f25f0f3c19
|
3 |
+
size 842230989
|