Hugging Face's logo

GregaSustar
/
codi_MCAC

Model card Files
xet
 Community
codi_MCAC
File size: 4,046 Bytes
cc1aa73
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
###################################################   Variables   ##################################################
define: &learn_sigma True

###################################################    Config     ##################################################
name: mcac4

vae:
  enabled: True
  path: "stabilityai/sdxl-vae"
  single_ch: False

diffusion:
  type: "Denoise"
  params:
    diffusion_steps: 1000
    noise_schedule: scaled_linear
    learn_sigma: *learn_sigma
    sigma_small: False
    use_kl: False
    parametrization: "v"
    rescale_timesteps: False
    rescale_learned_sigmas: False
    timestep_respacing: ""
    lmbd_vlb: 0.001
    lmbd_cb_count: 0.0
    t_mse_weighting_scheme: "uniform"
    t_cb_count_weighting_scheme: "uniform"
    enforce_zero_terminal_snr: True

model:
  type: "UNet"
  params:
    # input_size: 64
    in_channels: 164
    model_channels: 192
    out_channels: 4
    num_res_blocks: 2
    attention_resolutions: [1]
    dropout: 0.0
    channel_mult: [1, 2, 4]
    conv_resample: True
    dims: 2
    y_dim: null
    context_dim: 768
    use_checkpoint: True
    num_heads: -1
    num_head_channels: 64
    num_heads_upsample: -1
    use_scale_shift_norm: True
    resblock_updown: True
    learn_count: False
    learn_sigma: *learn_sigma
    transformer_depth: 1
    initial_ds: 0.125
    st_roi_encode_size: False
    st_roi_output_size: 1
    st_unfold_roi: False
    disable_middle_transformer: True
    enhanced_spatial_transformer: True
    nx_enhanced: 1
    disable_self_attentions: null
    deformable_self_attn: False
    topk: -1

conditioner:
  embedders:
    # - type: "ImageConcatEmbedder"
    #   input_keys: ["img"]
    #   is_trainable: False
    #   reference_key: "z_img"

    # - type: "DINOv2ImageEmbedder"
    #   input_keys: ["img"]
    #   is_trainable: True
    #   reference_key: "z_img"
    #   # custom_outkey: discard
    #   params:
    #     checkpoint: "facebook/dinov2-large"
    #     out_channels: 160

    - type: "RADIOImageEmbedder"
      input_keys: ["img"]
      is_trainable: True
      reference_key: "z_img"
      params:
        model_version: "radio_v2.5-l"
        out_channels: 160
        resize: True
        # out_indices: null

    # - type: "ProjectEmbedder"
    #   input_keys: ["z_img"]
    #   is_trainable: True
    #   params:
    #     in_channels: 256
    #     out_channels: 16

    - type: "BBoxAppendEmbedder"
      input_keys: ["bboxes"]
      is_trainable: False
      custom_outkey: "bboxes"
      custom_catdim: 1
      ucg_rate: 0.0
    
    - type: "BBoxSizeEmbedder"
      input_keys: ["bboxes"]
      is_trainable: True
      ucg_rate: 0.0
      params:
        outdim: 768

    # - type: "RoIAlignExemplarEmbedder"
    #   input_keys: ["z_img", "bboxes"]
    #   is_trainable: True
    #   ucg_rate: 0.0
    #   params:
    #     in_channels: 16
    #     out_channels: 768
    #     roi_output_size: 7
    #     remove_sequence_dim: False

    # - type: "FrozenCLIPEmbedder"
    #   is_trainable: True
    #   input_keys: ["text"]
    #   ucg_rate: 0.0
    #   params:
    #     out_channels: 768
    #     layer: hidden
    #     layer_idx: 11
data:
  dataset:
    name: "MCAC"
    params:
      datadir: /d/hpc/projects/FRI/DL/gs1121/MCAC/
      n_exemplars: 3
      image_size: 512
      occ_limit: 70
      M1: False
      hflip_p: 0.5
      vflip_p: 0.0
      mosaic_p: 0.5
      mosaic_avgbbox_h_lim: 50
      mosaic_avgbbox_w_lim: 50
      mosaic_avgbbox_area_lim: 720
      sigma: 0.5
      target_minmax_norm: True
      tlrb_center_sample: False
      tlrb_point_sample: False
      tlrb_radius: 8
      with_tlrb: False

  dataloader:
    params:
      batch_size: 16
      num_workers: 8

train:
  lr: 1.0e-4
  num_epochs: 200
  grad_clip: 1.0
  ema_rate: 0.9999
  validation_interval: 10
  resume_checkpoint: null
  weight_decay: 0.0001
  lr_scheduler: warmup_cosine_anneal
  overfit: False
  mixed_precision: False
  seed: null

log:
  logdir: ./logs/
  wandb_mode: online
  log_interval: 100
  save_interval: 50