Paolo-Fraccaro
commited on
Commit
•
91b09fa
1
Parent(s):
44130b3
Update burn_scars_Prithvi_100M.py
Browse files- burn_scars_Prithvi_100M.py +89 -183
burn_scars_Prithvi_100M.py
CHANGED
@@ -1,11 +1,17 @@
|
|
|
|
|
|
|
|
1 |
dist_params = dict(backend='nccl')
|
2 |
log_level = 'INFO'
|
3 |
load_from = None
|
4 |
resume_from = None
|
5 |
cudnn_benchmark = True
|
6 |
-
|
7 |
dataset_type = 'GeospatialDataset'
|
8 |
-
|
|
|
|
|
|
|
9 |
num_frames = 1
|
10 |
img_size = 224
|
11 |
num_workers = 4
|
@@ -22,45 +28,45 @@ img_norm_cfg = dict(
|
|
22 |
bands = [0, 1, 2, 3, 4, 5]
|
23 |
tile_size = 224
|
24 |
orig_nsize = 512
|
25 |
-
crop_size = (
|
26 |
img_suffix = '_merged.tif'
|
27 |
seg_map_suffix = '.mask.tif'
|
28 |
ignore_index = -1
|
29 |
image_nodata = -9999
|
30 |
image_nodata_replace = 0
|
31 |
image_to_float32 = True
|
32 |
-
|
33 |
-
|
|
|
|
|
34 |
num_layers = 12
|
35 |
patch_size = 16
|
36 |
embed_dim = 768
|
37 |
num_heads = 12
|
38 |
tubelet_size = 1
|
39 |
-
|
40 |
-
|
41 |
-
|
42 |
-
|
43 |
-
|
44 |
-
|
|
|
|
|
|
|
|
|
|
|
45 |
train_pipeline = [
|
46 |
-
dict(type='LoadGeospatialImageFromFile', to_float32=True),
|
47 |
dict(type='LoadGeospatialAnnotations', reduce_zero_label=False),
|
48 |
-
dict(type='BandsExtract', bands=
|
49 |
dict(type='RandomFlip', prob=0.5),
|
50 |
dict(type='ToTensor', keys=['img', 'gt_semantic_seg']),
|
51 |
-
|
52 |
-
|
53 |
-
|
54 |
-
|
55 |
-
|
56 |
-
|
57 |
-
stds=[
|
58 |
-
0.02269135568823774, 0.026807560223070237, 0.04004109844362779,
|
59 |
-
0.07791732423672691, 0.08708738838140137, 0.07241979477437814
|
60 |
-
]),
|
61 |
-
dict(type='TorchRandomCrop', crop_size=(224, 224)),
|
62 |
-
dict(type='Reshape', keys=['img'], new_shape=(6, 1, 224, 224)),
|
63 |
-
dict(type='Reshape', keys=['gt_semantic_seg'], new_shape=(1, 224, 224)),
|
64 |
dict(
|
65 |
type='CastTensor',
|
66 |
keys=['gt_semantic_seg'],
|
@@ -68,23 +74,16 @@ train_pipeline = [
|
|
68 |
dict(type='Collect', keys=['img', 'gt_semantic_seg'])
|
69 |
]
|
70 |
test_pipeline = [
|
71 |
-
dict(type='LoadGeospatialImageFromFile', to_float32=True),
|
72 |
-
dict(type='BandsExtract', bands=
|
73 |
dict(type='ToTensor', keys=['img']),
|
74 |
-
|
75 |
-
|
76 |
-
|
77 |
-
0.033349706741586264, 0.05701185520536176, 0.05889748132001316,
|
78 |
-
0.2323245113436119, 0.1972854853760658, 0.11944914225186566
|
79 |
-
],
|
80 |
-
stds=[
|
81 |
-
0.02269135568823774, 0.026807560223070237, 0.04004109844362779,
|
82 |
-
0.07791732423672691, 0.08708738838140137, 0.07241979477437814
|
83 |
-
]),
|
84 |
dict(
|
85 |
type='Reshape',
|
86 |
keys=['img'],
|
87 |
-
new_shape=(
|
88 |
look_up=dict({
|
89 |
'2': 1,
|
90 |
'3': 2
|
@@ -99,136 +98,43 @@ test_pipeline = [
|
|
99 |
'scale_factor', 'img_norm_cfg'
|
100 |
])
|
101 |
]
|
|
|
|
|
|
|
102 |
data = dict(
|
103 |
-
samples_per_gpu=
|
104 |
-
workers_per_gpu=
|
105 |
train=dict(
|
106 |
-
type=
|
107 |
-
|
108 |
-
|
109 |
img_dir='training',
|
110 |
ann_dir='training',
|
111 |
-
img_suffix=
|
112 |
-
seg_map_suffix=
|
113 |
-
pipeline=
|
114 |
-
dict(type='LoadGeospatialImageFromFile', to_float32=True),
|
115 |
-
dict(type='LoadGeospatialAnnotations', reduce_zero_label=False),
|
116 |
-
dict(type='BandsExtract', bands=[0, 1, 2, 3, 4, 5]),
|
117 |
-
dict(type='RandomFlip', prob=0.5),
|
118 |
-
dict(type='ToTensor', keys=['img', 'gt_semantic_seg']),
|
119 |
-
dict(
|
120 |
-
type='TorchNormalize',
|
121 |
-
means=[
|
122 |
-
0.033349706741586264, 0.05701185520536176,
|
123 |
-
0.05889748132001316, 0.2323245113436119,
|
124 |
-
0.1972854853760658, 0.11944914225186566
|
125 |
-
],
|
126 |
-
stds=[
|
127 |
-
0.02269135568823774, 0.026807560223070237,
|
128 |
-
0.04004109844362779, 0.07791732423672691,
|
129 |
-
0.08708738838140137, 0.07241979477437814
|
130 |
-
]),
|
131 |
-
dict(type='TorchRandomCrop', crop_size=(224, 224)),
|
132 |
-
dict(type='Reshape', keys=['img'], new_shape=(6, 1, 224, 224)),
|
133 |
-
dict(
|
134 |
-
type='Reshape',
|
135 |
-
keys=['gt_semantic_seg'],
|
136 |
-
new_shape=(1, 224, 224)),
|
137 |
-
dict(
|
138 |
-
type='CastTensor',
|
139 |
-
keys=['gt_semantic_seg'],
|
140 |
-
new_type='torch.LongTensor'),
|
141 |
-
dict(type='Collect', keys=['img', 'gt_semantic_seg'])
|
142 |
-
],
|
143 |
ignore_index=-1),
|
144 |
val=dict(
|
145 |
-
type=
|
146 |
-
|
147 |
-
|
148 |
img_dir='validation',
|
149 |
ann_dir='validation',
|
150 |
-
img_suffix=
|
151 |
-
seg_map_suffix=
|
152 |
-
pipeline=
|
153 |
-
dict(type='LoadGeospatialImageFromFile', to_float32=True),
|
154 |
-
dict(type='BandsExtract', bands=[0, 1, 2, 3, 4, 5]),
|
155 |
-
dict(type='ToTensor', keys=['img']),
|
156 |
-
dict(
|
157 |
-
type='TorchNormalize',
|
158 |
-
means=[
|
159 |
-
0.033349706741586264, 0.05701185520536176,
|
160 |
-
0.05889748132001316, 0.2323245113436119,
|
161 |
-
0.1972854853760658, 0.11944914225186566
|
162 |
-
],
|
163 |
-
stds=[
|
164 |
-
0.02269135568823774, 0.026807560223070237,
|
165 |
-
0.04004109844362779, 0.07791732423672691,
|
166 |
-
0.08708738838140137, 0.07241979477437814
|
167 |
-
]),
|
168 |
-
dict(
|
169 |
-
type='Reshape',
|
170 |
-
keys=['img'],
|
171 |
-
new_shape=(6, 1, -1, -1),
|
172 |
-
look_up=dict({
|
173 |
-
'2': 1,
|
174 |
-
'3': 2
|
175 |
-
})),
|
176 |
-
dict(
|
177 |
-
type='CastTensor', keys=['img'], new_type='torch.FloatTensor'),
|
178 |
-
dict(
|
179 |
-
type='CollectTestList',
|
180 |
-
keys=['img'],
|
181 |
-
meta_keys=[
|
182 |
-
'img_info', 'seg_fields', 'img_prefix', 'seg_prefix',
|
183 |
-
'filename', 'ori_filename', 'img', 'img_shape',
|
184 |
-
'ori_shape', 'pad_shape', 'scale_factor', 'img_norm_cfg'
|
185 |
-
])
|
186 |
-
],
|
187 |
ignore_index=-1),
|
188 |
test=dict(
|
189 |
-
type=
|
190 |
-
|
191 |
-
|
192 |
img_dir='validation',
|
193 |
ann_dir='validation',
|
194 |
-
img_suffix=
|
195 |
-
seg_map_suffix=
|
196 |
-
pipeline=
|
197 |
-
dict(type='LoadGeospatialImageFromFile', to_float32=True),
|
198 |
-
dict(type='BandsExtract', bands=[0, 1, 2, 3, 4, 5]),
|
199 |
-
dict(type='ToTensor', keys=['img']),
|
200 |
-
dict(
|
201 |
-
type='TorchNormalize',
|
202 |
-
means=[
|
203 |
-
0.033349706741586264, 0.05701185520536176,
|
204 |
-
0.05889748132001316, 0.2323245113436119,
|
205 |
-
0.1972854853760658, 0.11944914225186566
|
206 |
-
],
|
207 |
-
stds=[
|
208 |
-
0.02269135568823774, 0.026807560223070237,
|
209 |
-
0.04004109844362779, 0.07791732423672691,
|
210 |
-
0.08708738838140137, 0.07241979477437814
|
211 |
-
]),
|
212 |
-
dict(
|
213 |
-
type='Reshape',
|
214 |
-
keys=['img'],
|
215 |
-
new_shape=(6, 1, -1, -1),
|
216 |
-
look_up=dict({
|
217 |
-
'2': 1,
|
218 |
-
'3': 2
|
219 |
-
})),
|
220 |
-
dict(
|
221 |
-
type='CastTensor', keys=['img'], new_type='torch.FloatTensor'),
|
222 |
-
dict(
|
223 |
-
type='CollectTestList',
|
224 |
-
keys=['img'],
|
225 |
-
meta_keys=[
|
226 |
-
'img_info', 'seg_fields', 'img_prefix', 'seg_prefix',
|
227 |
-
'filename', 'ori_filename', 'img', 'img_shape',
|
228 |
-
'ori_shape', 'pad_shape', 'scale_factor', 'img_norm_cfg'
|
229 |
-
])
|
230 |
-
],
|
231 |
ignore_index=-1))
|
|
|
232 |
optimizer = dict(type='Adam', lr=1.3e-05, betas=(0.9, 0.999))
|
233 |
optimizer_config = dict(grad_clip=None)
|
234 |
lr_config = dict(
|
@@ -248,16 +154,20 @@ log_config = dict(
|
|
248 |
checkpoint_config = dict(
|
249 |
by_epoch=True,
|
250 |
interval=10,
|
251 |
-
out_dir=
|
252 |
-
'/dccstor/geofm-finetuning/carlosgomes/fire_scars/carlos_replicate_experiment_fixed_lr'
|
253 |
)
|
254 |
evaluation = dict(
|
255 |
-
interval=
|
256 |
metric='mIoU',
|
257 |
pre_eval=True,
|
258 |
save_best='mIoU',
|
259 |
by_epoch=False)
|
260 |
-
|
|
|
|
|
|
|
|
|
|
|
261 |
workflow = [('train', 1)]
|
262 |
norm_cfg = dict(type='BN', requires_grad=True)
|
263 |
model = dict(
|
@@ -265,28 +175,27 @@ model = dict(
|
|
265 |
frozen_backbone=False,
|
266 |
backbone=dict(
|
267 |
type='TemporalViTEncoder',
|
268 |
-
pretrained=
|
269 |
-
|
270 |
-
|
271 |
-
|
272 |
-
|
273 |
-
|
274 |
-
|
275 |
-
embed_dim=768,
|
276 |
depth=12,
|
277 |
-
num_heads=
|
278 |
mlp_ratio=4.0,
|
279 |
norm_pix_loss=False),
|
280 |
neck=dict(
|
281 |
type='ConvTransformerTokensToEmbeddingNeck',
|
282 |
-
embed_dim=
|
283 |
-
output_embed_dim=
|
284 |
drop_cls_token=True,
|
285 |
Hp=14,
|
286 |
Wp=14),
|
287 |
decode_head=dict(
|
288 |
-
num_classes=
|
289 |
-
in_channels=
|
290 |
type='FCNHead',
|
291 |
in_index=-1,
|
292 |
channels=256,
|
@@ -295,12 +204,11 @@ model = dict(
|
|
295 |
dropout_ratio=0.1,
|
296 |
norm_cfg=dict(type='BN', requires_grad=True),
|
297 |
align_corners=False,
|
298 |
-
loss_decode=
|
299 |
-
|
300 |
-
ignore_index=-1)),
|
301 |
auxiliary_head=dict(
|
302 |
-
num_classes=
|
303 |
-
in_channels=
|
304 |
type='FCNHead',
|
305 |
in_index=-1,
|
306 |
channels=256,
|
@@ -309,10 +217,8 @@ model = dict(
|
|
309 |
dropout_ratio=0.1,
|
310 |
norm_cfg=dict(type='BN', requires_grad=True),
|
311 |
align_corners=False,
|
312 |
-
loss_decode=
|
313 |
-
type='DiceLoss', use_sigmoid=False, loss_weight=1,
|
314 |
-
ignore_index=-1)),
|
315 |
train_cfg=dict(),
|
316 |
-
test_cfg=dict(mode='slide', stride=(
|
317 |
gpu_ids = range(0, 1)
|
318 |
-
auto_resume = False
|
|
|
1 |
+
import os
|
2 |
+
custom_imports = dict(imports=['geospatial_fm'])
|
3 |
+
|
4 |
dist_params = dict(backend='nccl')
|
5 |
log_level = 'INFO'
|
6 |
load_from = None
|
7 |
resume_from = None
|
8 |
cudnn_benchmark = True
|
9 |
+
|
10 |
dataset_type = 'GeospatialDataset'
|
11 |
+
|
12 |
+
# TO BE DEFINED BY USER: data directory
|
13 |
+
data_root = '<path to data root>'
|
14 |
+
|
15 |
num_frames = 1
|
16 |
img_size = 224
|
17 |
num_workers = 4
|
|
|
28 |
bands = [0, 1, 2, 3, 4, 5]
|
29 |
tile_size = 224
|
30 |
orig_nsize = 512
|
31 |
+
crop_size = (tile_size, tile_size)
|
32 |
img_suffix = '_merged.tif'
|
33 |
seg_map_suffix = '.mask.tif'
|
34 |
ignore_index = -1
|
35 |
image_nodata = -9999
|
36 |
image_nodata_replace = 0
|
37 |
image_to_float32 = True
|
38 |
+
|
39 |
+
# model
|
40 |
+
# TO BE DEFINED BY USER: model path
|
41 |
+
pretrained_weights_path = '<path to pretrained weights>'
|
42 |
num_layers = 12
|
43 |
patch_size = 16
|
44 |
embed_dim = 768
|
45 |
num_heads = 12
|
46 |
tubelet_size = 1
|
47 |
+
output_embed_dim = num_frames*embed_dim
|
48 |
+
max_intervals=10000
|
49 |
+
evaluation_interval=1000
|
50 |
+
|
51 |
+
# TO BE DEFINED BY USER: model path
|
52 |
+
experiment = '<experiment name>'
|
53 |
+
project_dir = '<project directory name>'
|
54 |
+
work_dir = os.path.join(project_dir, experiment)
|
55 |
+
save_path = work_dir
|
56 |
+
|
57 |
+
save_path = work_dir
|
58 |
train_pipeline = [
|
59 |
+
dict(type='LoadGeospatialImageFromFile', to_float32=image_to_float32, channels_last=True),
|
60 |
dict(type='LoadGeospatialAnnotations', reduce_zero_label=False),
|
61 |
+
dict(type='BandsExtract', bands=bands),
|
62 |
dict(type='RandomFlip', prob=0.5),
|
63 |
dict(type='ToTensor', keys=['img', 'gt_semantic_seg']),
|
64 |
+
# to channels first
|
65 |
+
dict(type="TorchPermute", keys=["img"], order=(2, 0, 1)),
|
66 |
+
dict(type='TorchNormalize', **img_norm_cfg),
|
67 |
+
dict(type='TorchRandomCrop', crop_size=(tile_size, tile_size)),
|
68 |
+
dict(type='Reshape', keys=['img'], new_shape=(len(bands), num_frames, tile_size, tile_size)),
|
69 |
+
dict(type='Reshape', keys=['gt_semantic_seg'], new_shape=(1, tile_size, tile_size)),
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
70 |
dict(
|
71 |
type='CastTensor',
|
72 |
keys=['gt_semantic_seg'],
|
|
|
74 |
dict(type='Collect', keys=['img', 'gt_semantic_seg'])
|
75 |
]
|
76 |
test_pipeline = [
|
77 |
+
dict(type='LoadGeospatialImageFromFile', to_float32=image_to_float32, channels_last=True),
|
78 |
+
dict(type='BandsExtract', bands=bands),
|
79 |
dict(type='ToTensor', keys=['img']),
|
80 |
+
# to channels first
|
81 |
+
dict(type="TorchPermute", keys=["img"], order=(2, 0, 1)),
|
82 |
+
dict(type='TorchNormalize', **img_norm_cfg),
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
83 |
dict(
|
84 |
type='Reshape',
|
85 |
keys=['img'],
|
86 |
+
new_shape=(len(bands), num_frames, -1, -1),
|
87 |
look_up=dict({
|
88 |
'2': 1,
|
89 |
'3': 2
|
|
|
98 |
'scale_factor', 'img_norm_cfg'
|
99 |
])
|
100 |
]
|
101 |
+
|
102 |
+
CLASSES = ('Unburnt land', 'Burn scar')
|
103 |
+
|
104 |
data = dict(
|
105 |
+
samples_per_gpu=samples_per_gpu,
|
106 |
+
workers_per_gpu=num_workers,
|
107 |
train=dict(
|
108 |
+
type=dataset_type,
|
109 |
+
CLASSES=CLASSES,
|
110 |
+
data_root=data_root,
|
111 |
img_dir='training',
|
112 |
ann_dir='training',
|
113 |
+
img_suffix=img_suffix,
|
114 |
+
seg_map_suffix=seg_map_suffix,
|
115 |
+
pipeline=train_pipeline,
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
116 |
ignore_index=-1),
|
117 |
val=dict(
|
118 |
+
type=dataset_type,
|
119 |
+
CLASSES=CLASSES,
|
120 |
+
data_root=data_root,
|
121 |
img_dir='validation',
|
122 |
ann_dir='validation',
|
123 |
+
img_suffix=img_suffix,
|
124 |
+
seg_map_suffix=seg_map_suffix,
|
125 |
+
pipeline=test_pipeline,
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
126 |
ignore_index=-1),
|
127 |
test=dict(
|
128 |
+
type=dataset_type,
|
129 |
+
CLASSES=CLASSES,
|
130 |
+
data_root=data_root,
|
131 |
img_dir='validation',
|
132 |
ann_dir='validation',
|
133 |
+
img_suffix=img_suffix,
|
134 |
+
seg_map_suffix=seg_map_suffix,
|
135 |
+
pipeline=test_pipeline,
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
136 |
ignore_index=-1))
|
137 |
+
|
138 |
optimizer = dict(type='Adam', lr=1.3e-05, betas=(0.9, 0.999))
|
139 |
optimizer_config = dict(grad_clip=None)
|
140 |
lr_config = dict(
|
|
|
154 |
checkpoint_config = dict(
|
155 |
by_epoch=True,
|
156 |
interval=10,
|
157 |
+
out_dir=save_path
|
|
|
158 |
)
|
159 |
evaluation = dict(
|
160 |
+
interval=evaluation_interval,
|
161 |
metric='mIoU',
|
162 |
pre_eval=True,
|
163 |
save_best='mIoU',
|
164 |
by_epoch=False)
|
165 |
+
|
166 |
+
loss_func=dict(
|
167 |
+
type='DiceLoss', use_sigmoid=False, loss_weight=1,
|
168 |
+
ignore_index=-1)
|
169 |
+
|
170 |
+
runner = dict(type='IterBasedRunner', max_iters=max_intervals)
|
171 |
workflow = [('train', 1)]
|
172 |
norm_cfg = dict(type='BN', requires_grad=True)
|
173 |
model = dict(
|
|
|
175 |
frozen_backbone=False,
|
176 |
backbone=dict(
|
177 |
type='TemporalViTEncoder',
|
178 |
+
pretrained=pretrained_weights_path,
|
179 |
+
img_size=img_size,
|
180 |
+
patch_size=patch_size,
|
181 |
+
num_frames=num_frames,
|
182 |
+
tubelet_size=tubelet_size,
|
183 |
+
in_chans=len(bands),
|
184 |
+
embed_dim=embed_dim,
|
|
|
185 |
depth=12,
|
186 |
+
num_heads=num_heads,
|
187 |
mlp_ratio=4.0,
|
188 |
norm_pix_loss=False),
|
189 |
neck=dict(
|
190 |
type='ConvTransformerTokensToEmbeddingNeck',
|
191 |
+
embed_dim=embed_dim*num_frames,
|
192 |
+
output_embed_dim=output_embed_dim,
|
193 |
drop_cls_token=True,
|
194 |
Hp=14,
|
195 |
Wp=14),
|
196 |
decode_head=dict(
|
197 |
+
num_classes=len(CLASSES),
|
198 |
+
in_channels=output_embed_dim,
|
199 |
type='FCNHead',
|
200 |
in_index=-1,
|
201 |
channels=256,
|
|
|
204 |
dropout_ratio=0.1,
|
205 |
norm_cfg=dict(type='BN', requires_grad=True),
|
206 |
align_corners=False,
|
207 |
+
loss_decode=
|
208 |
+
loss_decode=loss_func),
|
|
|
209 |
auxiliary_head=dict(
|
210 |
+
num_classes=len(CLASSES),
|
211 |
+
in_channels=output_embed_dim,
|
212 |
type='FCNHead',
|
213 |
in_index=-1,
|
214 |
channels=256,
|
|
|
217 |
dropout_ratio=0.1,
|
218 |
norm_cfg=dict(type='BN', requires_grad=True),
|
219 |
align_corners=False,
|
220 |
+
loss_decode=loss_func),
|
|
|
|
|
221 |
train_cfg=dict(),
|
222 |
+
test_cfg=dict(mode='slide', stride=(tile_size/2, tile_size/2), crop_size=(tile_size, tile_size)))
|
223 |
gpu_ids = range(0, 1)
|
224 |
+
auto_resume = False
|