Spaces:
Paused
Paused
Upload EVA02.yaml
Browse files- GLEE/configs/EVA02.yaml +72 -0
GLEE/configs/EVA02.yaml
ADDED
@@ -0,0 +1,72 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
MODEL:
|
2 |
+
META_ARCHITECTURE: "GLEE"
|
3 |
+
MASK_ON: True
|
4 |
+
BACKBONE:
|
5 |
+
NAME: "D2_EVA02"
|
6 |
+
EVA02:
|
7 |
+
CHECKPOINT: False
|
8 |
+
PIXEL_MEAN: [123.675, 116.280, 103.530]
|
9 |
+
PIXEL_STD: [58.395, 57.120, 57.375]
|
10 |
+
RESNETS:
|
11 |
+
DEPTH: 50
|
12 |
+
STEM_TYPE: "basic" # not used
|
13 |
+
STEM_OUT_CHANNELS: 64
|
14 |
+
STRIDE_IN_1X1: False
|
15 |
+
OUT_FEATURES: ["res2", "res3", "res4", "res5"]
|
16 |
+
# NORM: "SyncBN"
|
17 |
+
RES5_MULTI_GRID: [1, 1, 1] # not used
|
18 |
+
SEM_SEG_HEAD:
|
19 |
+
PIXEL_DECODER_NAME: "MaskDINOEncoder"
|
20 |
+
DIM_FEEDFORWARD: 2048
|
21 |
+
NUM_FEATURE_LEVELS: 4
|
22 |
+
TOTAL_NUM_FEATURE_LEVELS: 4
|
23 |
+
IN_FEATURES: ["p3", "p4", "p5", "p6"]
|
24 |
+
DEFORMABLE_TRANSFORMER_ENCODER_IN_FEATURES: ["p3", "p4", "p5", "p6"]
|
25 |
+
COMMON_STRIDE: 4
|
26 |
+
TRANSFORMER_ENC_LAYERS: 6
|
27 |
+
FEATURE_ORDER: "low2high"
|
28 |
+
NAME: "MaskDINOHead"
|
29 |
+
IGNORE_VALUE: 255
|
30 |
+
NUM_CLASSES: 80
|
31 |
+
LOSS_WEIGHT: 1.0
|
32 |
+
CONVS_DIM: 256
|
33 |
+
MASK_DIM: 256
|
34 |
+
NORM: "GN"
|
35 |
+
# pixel decoder
|
36 |
+
MaskDINO:
|
37 |
+
TRANSFORMER_DECODER_NAME: "MaskDINODecoder"
|
38 |
+
DEEP_SUPERVISION: True
|
39 |
+
NO_OBJECT_WEIGHT: 0.1
|
40 |
+
CLASS_WEIGHT: 4.0
|
41 |
+
MASK_WEIGHT: 5.0
|
42 |
+
DICE_WEIGHT: 5.0
|
43 |
+
BOX_WEIGHT: 5.0
|
44 |
+
GIOU_WEIGHT: 2.0
|
45 |
+
HIDDEN_DIM: 256
|
46 |
+
NUM_OBJECT_QUERIES: 300
|
47 |
+
NHEADS: 8
|
48 |
+
DROPOUT: 0.0
|
49 |
+
DIM_FEEDFORWARD: 2048
|
50 |
+
ENC_LAYERS: 0
|
51 |
+
PRE_NORM: False
|
52 |
+
ENFORCE_INPUT_PROJ: False
|
53 |
+
SIZE_DIVISIBILITY: 32
|
54 |
+
DEC_LAYERS: 9 # 9+1, 9 decoder layers, add one for the loss on learnable query
|
55 |
+
TRAIN_NUM_POINTS: 12544
|
56 |
+
OVERSAMPLE_RATIO: 3.0
|
57 |
+
IMPORTANCE_SAMPLE_RATIO: 0.75
|
58 |
+
INITIAL_PRED: True
|
59 |
+
TWO_STAGE: True
|
60 |
+
DN: "standard"
|
61 |
+
DN_NUM: 100
|
62 |
+
INITIALIZE_BOX_TYPE: "no"
|
63 |
+
TEST:
|
64 |
+
SEMANTIC_ON: False
|
65 |
+
INSTANCE_ON: True
|
66 |
+
PANOPTIC_ON: False
|
67 |
+
OVERLAP_THRESHOLD: 0.8
|
68 |
+
OBJECT_MASK_THRESHOLD: 0.25
|
69 |
+
TEXT:
|
70 |
+
ARCH: clip_teacher
|
71 |
+
LANGUAGE_BACKBONE:
|
72 |
+
LANG_DIM: 512
|