File size: 4,537 Bytes
edebe10
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
# Define Test/Trainer/Saving
PIPELINE: XDecoderPipeline
TRAINER: xdecoder
SAVE_DIR: '../../data/output/test'
base_path: "./"

# Resume Logistic
RESUME: false
WEIGHT: false
RESUME_FROM: ''
EVAL_AT_START: false

# Logging and Debug
WANDB: False
LOG_EVERY: 100
FIND_UNUSED_PARAMETERS: false

# Speed up training
FP16: false
PORT: '36873'

# misc
LOADER:
  JOINT: False
  KEY_DATASET: 'coco'

STANDARD_TEXT_FOR_EVAL: False

##################
# Task settings
##################
VERBOSE: true
MODEL:
  NAME: seem_model_demo
  HEAD: xdecoder_head
  DIM_PROJ: 512
  TEXT:
    ARCH: vlpencoder
    NAME: transformer
    TOKENIZER: clip
    CONTEXT_LENGTH: 77 # 77
    WIDTH: 512
    HEADS: 8
    LAYERS: 12 # 6
    AUTOGRESSIVE: True
  BACKBONE:
    NAME: focal
    PRETRAINED: ''
    LOAD_PRETRAINED: false
    FOCAL:
      PRETRAIN_IMG_SIZE: 224
      PATCH_SIZE: 4
      EMBED_DIM: 192
      DEPTHS: [2, 2, 18, 2]
      FOCAL_LEVELS: [4, 4, 4, 4]
      FOCAL_WINDOWS: [3, 3, 3, 3]
      DROP_PATH_RATE: 0.3
      MLP_RATIO: 4.0
      DROP_RATE: 0.0
      PATCH_NORM: True
      USE_CONV_EMBED: True
      SCALING_MODULATOR: True
      USE_CHECKPOINT: False
      USE_POSTLN: true
      USE_POSTLN_IN_MODULATION: false
      USE_LAYERSCALE: True
      OUT_FEATURES: ["res2", "res3", "res4", "res5"]
      OUT_INDICES: [0, 1, 2, 3]
  ENCODER:
    NAME: transformer_encoder_fpn
    IGNORE_VALUE: 255
    NUM_CLASSES: 16
    BINARY_CLASSES: False
    LOSS_WEIGHT: 1.0
    CONVS_DIM: 512
    MASK_DIM: 512
    NORM: "GN"
    IN_FEATURES: ["res2", "res3", "res4", "res5"]
    DEFORMABLE_TRANSFORMER_ENCODER_IN_FEATURES: ["res3", "res4", "res5"]
    COMMON_STRIDE: 4
    TRANSFORMER_ENC_LAYERS: 6
  DECODER:
    NAME: seem_demo
    TRANSFORMER_IN_FEATURE: "multi_scale_pixel_decoder"
    MASK:
      ENABLED: False
    DETECTION: False
    SPATIAL:
      ENABLED: True
      MAX_ITER: 1
    GROUNDING:
      ENABLED: True
      MAX_LEN: 5
      TEXT_WEIGHT: 2.0
      CLASS_WEIGHT: 0.5
    VISUAL:
      ENABLED: False
    AUDIO:
      ENABLED: False
    RETRIEVAL:
      ENABLED: False
    LVIS:
      ENABLED: True
      THRES: 0.7
    OPENIMAGE:
      ENABLED: False
      NEGATIVE_SAMPLES: 5
      GROUNDING:
        ENABLED: False
        MAX_LEN: 5
    CAPTION:
      ENABLED: False
      PHRASE_PROB: 0.5
      SIM_THRES: 0.95
    DEEP_SUPERVISION: True
    NO_OBJECT_WEIGHT: 0.1
    GCLASS_WEIGHT: 0.4
    GMASK_WEIGHT: 1.0
    GDICE_WEIGHT: 1.0
    SCLASS_WEIGHT: 0.4
    SMASK_WEIGHT: 1.0
    SDICE_WEIGHT: 1.0
    OCLASS_WEIGHT: 0.4
    OMASK_WEIGHT: 1.0
    ODICE_WEIGHT: 1.0
    CLASS_WEIGHT: 2.0
    MASK_WEIGHT: 5.0
    DICE_WEIGHT: 5.0
    BBOX_WEIGHT: 5.0
    GIOU_WEIGHT: 2.0
    CAPTION_WEIGHT: 2.0
    COST_SPATIAL:
      CLASS_WEIGHT: 5.0
      MASK_WEIGHT: 2.0
      DICE_WEIGHT: 2.0
    HIDDEN_DIM: 512
    NUM_OBJECT_QUERIES: 101
    NHEADS: 8
    DROPOUT: 0.0
    DIM_FEEDFORWARD: 2048
    MAX_SPATIAL_LEN: [512, 512, 512, 512]
    # ENC_LAYERS: 0
    PRE_NORM: False
    ENFORCE_INPUT_PROJ: False
    SIZE_DIVISIBILITY: 32
    TRAIN_NUM_POINTS: 12544
    OVERSAMPLE_RATIO: 3.0
    IMPORTANCE_SAMPLE_RATIO: 0.75
    DEC_LAYERS: 10  # 9 decoder layers, add one for the loss on learnable query
    TOP_GROUNDING_LAYERS: 10
    TOP_CAPTION_LAYERS: 10
    TOP_SPATIAL_LAYERS: 10
    TOP_OPENIMAGE_LAYERS: 10
    TEST:
      SEMANTIC_ON: True
      INSTANCE_ON: True
      PANOPTIC_ON: True
      OVERLAP_THRESHOLD: 0.8
      OBJECT_MASK_THRESHOLD: 0.4
      SEM_SEG_POSTPROCESSING_BEFORE_INFERENCE: false
      DETECTIONS_PER_IMAGE: 100

# Multi-modal Architecture, order matters
ATTENTION_ARCH:
  VARIABLE:
    queries: ['object']
    tokens: ['grounding', 'spatial', 'visual', 'audio']
  SELF_ATTENTION:
    queries:
      object: ['queries_object', 'tokens_grounding', 'tokens_spatial', 'tokens_visual', 'tokens_audio']
    tokens:
      grounding: ['queries_object', 'tokens_grounding']
      spatial: ['tokens_spatial']
      visual: ['tokens_visual']
      audio: ['queries_object', 'tokens_audio']
  CROSS_ATTENTION:
    queries:
      object: True
    tokens:
      grounding: False
      spatial: False
      visual: False
      audio: False
  MASKING: ['tokens_spatial', 'tokens_grounding', 'tokens_visual', 'tokens_audio']
  DUPLICATION:
    queries:
      grounding: 'queries_object'
      spatial: 'queries_object'
  SPATIAL_MEMORIES: 32

INPUT:
  PIXEL_MEAN: [123.675, 116.280, 103.530]
  PIXEL_STD: [58.395, 57.120, 57.375]
# INPUT:
#   PIXEL_MEAN: [64.284, 59.293, 59.962]
#   PIXEL_STD: [62.484, 60.865, 59.835]