File size: 10,812 Bytes
bd27f44
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
# Copyright (c) Facebook, Inc. and its affiliates.
# Copyright (c) Meta Platforms, Inc. All Rights Reserved
# Modified by Feng Liang from https://github.com/MendelXu/zsseg.baseline/blob/master/train_net.py

"""
OVSeg Training Script.

This script is a simplified version of the training script in detectron2/tools.
"""
import copy
import itertools
import logging
import os
from collections import OrderedDict
from typing import Any, Dict, List, Set

import detectron2.utils.comm as comm
import torch
from detectron2.checkpoint import DetectionCheckpointer
from detectron2.config import get_cfg
from detectron2.data import MetadataCatalog
from detectron2.engine import (
    DefaultTrainer,
    default_argument_parser,
    default_setup,
    launch,
)
from detectron2.evaluation import (
    DatasetEvaluator,
    CityscapesSemSegEvaluator,
    COCOEvaluator,
    DatasetEvaluators,
    verify_results,
)
from detectron2.projects.deeplab import add_deeplab_config, build_lr_scheduler
from detectron2.solver.build import maybe_add_gradient_clipping
from detectron2.utils.logger import setup_logger
from detectron2.utils.events import CommonMetricPrinter, JSONWriter

# OVSeg
from open_vocab_seg import SemanticSegmentorWithTTA, add_ovseg_config
from open_vocab_seg.data import (
    MaskFormerSemanticDatasetMapper,
)

from open_vocab_seg.data import (
    build_detection_test_loader,
    build_detection_train_loader,
)
from open_vocab_seg.evaluation import (
    GeneralizedSemSegEvaluator,
)
from open_vocab_seg.utils.events import WandbWriter, setup_wandb
from open_vocab_seg.utils.post_process_utils import dense_crf_post_process


class Trainer(DefaultTrainer):
    """
    Extension of the Trainer class adapted to DETR.
    """

    @classmethod
    def build_evaluator(cls, cfg, dataset_name, output_folder=None):
        """
        Create evaluator(s) for a given dataset.
        This uses the special metadata "evaluator_type" associated with each
        builtin dataset. For your own dataset, you can simply create an
        evaluator manually in your script and do not have to worry about the
        hacky if-else logic here.
        """
        if output_folder is None:
            output_folder = os.path.join(cfg.OUTPUT_DIR, "inference")
        evaluator_list = []
        evaluator_type = MetadataCatalog.get(dataset_name).evaluator_type
        if evaluator_type in ["sem_seg"]:
            evaluator = GeneralizedSemSegEvaluator
            evaluator_list.append(
                evaluator(
                    dataset_name,
                    distributed=True,
                    output_dir=output_folder,
                    post_process_func=dense_crf_post_process
                    if cfg.TEST.DENSE_CRF
                    else None,
                )
            )

        if len(evaluator_list) == 0:
            raise NotImplementedError(
                "no Evaluator for the dataset {} with the type {}".format(
                    dataset_name, evaluator_type
                )
            )
        elif len(evaluator_list) == 1:
            return evaluator_list[0]
        return DatasetEvaluators(evaluator_list)

    @classmethod
    def build_train_loader(cls, cfg):
        dataset = None
        # Semantic segmentation dataset mapper
        if cfg.INPUT.DATASET_MAPPER_NAME == "mask_former_semantic":
            mapper = MaskFormerSemanticDatasetMapper(cfg, True)
        else:
            raise NotImplementedError
        return build_detection_train_loader(cfg, mapper=mapper, dataset=dataset)

    @classmethod
    def build_test_loader(cls, cfg, dataset_name):
        """
        Returns:
            iterable
        It now calls :func:`detectron2.data.build_detection_test_loader`.
        Overwrite it if you'd like a different data loader.
        """
        return build_detection_test_loader(cfg, dataset_name, mapper=None)

    def build_writers(self):
        """
        Build a list of writers to be used. By default it contains
        writers that write metrics to the screen,
        a json file, and a tensorboard event file respectively.
        If you'd like a different list of writers, you can overwrite it in
        your trainer.

        Returns:
            list[EventWriter]: a list of :class:`EventWriter` objects.

        It is now implemented by:
        ::
            return [
                CommonMetricPrinter(self.max_iter),
                JSONWriter(os.path.join(self.cfg.OUTPUT_DIR, "metrics.json")),
                TensorboardXWriter(self.cfg.OUTPUT_DIR),
            ]

        """
        # Here the default print/log frequency of each writer is used.
        return [
            # It may not always print what you want to see, since it prints "common" metrics only.
            CommonMetricPrinter(self.max_iter),
            JSONWriter(os.path.join(self.cfg.OUTPUT_DIR, "metrics.json")),
            WandbWriter(),
        ]

    @classmethod
    def build_lr_scheduler(cls, cfg, optimizer):
        """
        It now calls :func:`detectron2.solver.build_lr_scheduler`.
        Overwrite it if you'd like a different scheduler.
        """
        return build_lr_scheduler(cfg, optimizer)

    @classmethod
    def build_optimizer(cls, cfg, model):
        weight_decay_norm = cfg.SOLVER.WEIGHT_DECAY_NORM
        weight_decay_embed = cfg.SOLVER.WEIGHT_DECAY_EMBED

        defaults = {}
        defaults["lr"] = cfg.SOLVER.BASE_LR
        defaults["weight_decay"] = cfg.SOLVER.WEIGHT_DECAY

        norm_module_types = (
            torch.nn.BatchNorm1d,
            torch.nn.BatchNorm2d,
            torch.nn.BatchNorm3d,
            torch.nn.SyncBatchNorm,
            # NaiveSyncBatchNorm inherits from BatchNorm2d
            torch.nn.GroupNorm,
            torch.nn.InstanceNorm1d,
            torch.nn.InstanceNorm2d,
            torch.nn.InstanceNorm3d,
            torch.nn.LayerNorm,
            torch.nn.LocalResponseNorm,
        )

        params: List[Dict[str, Any]] = []
        memo: Set[torch.nn.parameter.Parameter] = set()
        for module_name, module in model.named_modules():
            for module_param_name, value in module.named_parameters(recurse=False):
                if not value.requires_grad:
                    continue
                # Avoid duplicating parameters
                if value in memo:
                    continue
                memo.add(value)

                hyperparams = copy.copy(defaults)
                if "backbone" in module_name:
                    hyperparams["lr"] = (
                        hyperparams["lr"] * cfg.SOLVER.BACKBONE_MULTIPLIER
                    )
                if (
                    "relative_position_bias_table" in module_param_name
                    or "absolute_pos_embed" in module_param_name
                ):
                    print(module_param_name)
                    hyperparams["weight_decay"] = 0.0
                if isinstance(module, norm_module_types):
                    hyperparams["weight_decay"] = weight_decay_norm
                if isinstance(module, torch.nn.Embedding):
                    hyperparams["weight_decay"] = weight_decay_embed
                params.append({"params": [value], **hyperparams})

        def maybe_add_full_model_gradient_clipping(optim):
            # detectron2 doesn't have full model gradient clipping now
            clip_norm_val = cfg.SOLVER.CLIP_GRADIENTS.CLIP_VALUE
            enable = (
                cfg.SOLVER.CLIP_GRADIENTS.ENABLED
                and cfg.SOLVER.CLIP_GRADIENTS.CLIP_TYPE == "full_model"
                and clip_norm_val > 0.0
            )

            class FullModelGradientClippingOptimizer(optim):
                def step(self, closure=None):
                    all_params = itertools.chain(
                        *[x["params"] for x in self.param_groups]
                    )
                    torch.nn.utils.clip_grad_norm_(all_params, clip_norm_val)
                    super().step(closure=closure)

            return FullModelGradientClippingOptimizer if enable else optim

        optimizer_type = cfg.SOLVER.OPTIMIZER
        if optimizer_type == "SGD":
            optimizer = maybe_add_full_model_gradient_clipping(torch.optim.SGD)(
                params, cfg.SOLVER.BASE_LR, momentum=cfg.SOLVER.MOMENTUM
            )
        elif optimizer_type == "ADAMW":
            optimizer = maybe_add_full_model_gradient_clipping(torch.optim.AdamW)(
                params, cfg.SOLVER.BASE_LR
            )
        else:
            raise NotImplementedError(f"no optimizer type {optimizer_type}")
        if not cfg.SOLVER.CLIP_GRADIENTS.CLIP_TYPE == "full_model":
            optimizer = maybe_add_gradient_clipping(cfg, optimizer)
        return optimizer

    @classmethod
    def test_with_TTA(cls, cfg, model):
        logger = logging.getLogger("detectron2.trainer")
        # In the end of training, run an evaluation with TTA.
        logger.info("Running inference with test-time augmentation ...")
        model = SemanticSegmentorWithTTA(cfg, model)
        evaluators = [
            cls.build_evaluator(
                cfg, name, output_folder=os.path.join(cfg.OUTPUT_DIR, "inference_TTA")
            )
            for name in cfg.DATASETS.TEST
        ]
        res = cls.test(cfg, model, evaluators)
        res = OrderedDict({k + "_TTA": v for k, v in res.items()})
        return res


def setup(args):
    """
    Create configs and perform basic setups.
    """
    cfg = get_cfg()
    # for poly lr schedule
    add_deeplab_config(cfg)
    add_ovseg_config(cfg)
    cfg.merge_from_file(args.config_file)
    cfg.merge_from_list(args.opts)
    cfg.freeze()
    default_setup(cfg, args)
    # Setup logger for "ovseg" module
    if not args.eval_only:
        setup_wandb(cfg, args)
    setup_logger(
        output=cfg.OUTPUT_DIR, distributed_rank=comm.get_rank(), name="ovseg"
    )
    return cfg


def main(args):
    cfg = setup(args)

    if args.eval_only:
        model = Trainer.build_model(cfg)
        DetectionCheckpointer(model, save_dir=cfg.OUTPUT_DIR).resume_or_load(
            cfg.MODEL.WEIGHTS, resume=args.resume
        )

        if cfg.TEST.AUG.ENABLED:
            res = Trainer.test_with_TTA(cfg, model)
        else:
            res = Trainer.test(cfg, model)
        if comm.is_main_process():
            verify_results(cfg, res)
        return res

    trainer = Trainer(cfg)
    trainer.resume_or_load(resume=args.resume)
    return trainer.train()


if __name__ == "__main__":
    args = default_argument_parser().parse_args()
    print("Command Line Args:", args)
    launch(
        main,
        args.num_gpus,
        num_machines=args.num_machines,
        machine_rank=args.machine_rank,
        dist_url=args.dist_url,
        args=(args,),
    )