loliipopshock
commited on
Commit
·
e483cda
1
Parent(s):
f9b14aa
Add training scripts
Browse files- tools/train_net.py +150 -0
tools/train_net.py
ADDED
@@ -0,0 +1,150 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
"""
|
2 |
+
The script is based on https://github.com/facebookresearch/detectron2/blob/master/tools/train_net.py.
|
3 |
+
"""
|
4 |
+
|
5 |
+
import logging
|
6 |
+
import os
|
7 |
+
import json
|
8 |
+
from collections import OrderedDict
|
9 |
+
import torch
|
10 |
+
import sys
|
11 |
+
import detectron2.utils.comm as comm
|
12 |
+
from detectron2.checkpoint import DetectionCheckpointer
|
13 |
+
from detectron2.config import get_cfg
|
14 |
+
|
15 |
+
from detectron2.data import MetadataCatalog, DatasetCatalog
|
16 |
+
from detectron2.data.datasets import register_coco_instances
|
17 |
+
|
18 |
+
from detectron2.engine import DefaultTrainer, default_argument_parser, default_setup, hooks, launch
|
19 |
+
from detectron2.evaluation import (
|
20 |
+
COCOEvaluator,
|
21 |
+
DatasetEvaluators,
|
22 |
+
SemSegEvaluator,
|
23 |
+
verify_results,
|
24 |
+
)
|
25 |
+
from detectron2.modeling import GeneralizedRCNNWithTTA
|
26 |
+
import pandas as pd
|
27 |
+
|
28 |
+
class Trainer(DefaultTrainer):
|
29 |
+
"""
|
30 |
+
We use the "DefaultTrainer" which contains pre-defined default logic for
|
31 |
+
standard training workflow. They may not work for you, especially if you
|
32 |
+
are working on a new research project. In that case you can use the cleaner
|
33 |
+
"SimpleTrainer", or write your own training loop. You can use
|
34 |
+
"tools/plain_train_net.py" as an example.
|
35 |
+
"""
|
36 |
+
|
37 |
+
@classmethod
|
38 |
+
def build_evaluator(cls, cfg, dataset_name, output_folder=None):
|
39 |
+
"""
|
40 |
+
Returns:
|
41 |
+
DatasetEvaluator or None
|
42 |
+
|
43 |
+
It is not implemented by default.
|
44 |
+
"""
|
45 |
+
return COCOEvaluator(dataset_name, cfg, True, output_folder)
|
46 |
+
|
47 |
+
@classmethod
|
48 |
+
def test_with_TTA(cls, cfg, model):
|
49 |
+
logger = logging.getLogger("detectron2.trainer")
|
50 |
+
# In the end of training, run an evaluation with TTA
|
51 |
+
# Only support some R-CNN models.
|
52 |
+
logger.info("Running inference with test-time augmentation ...")
|
53 |
+
model = GeneralizedRCNNWithTTA(cfg, model)
|
54 |
+
evaluators = [
|
55 |
+
cls.build_evaluator(
|
56 |
+
cfg, name, output_folder=os.path.join(cfg.OUTPUT_DIR, "inference_TTA")
|
57 |
+
)
|
58 |
+
for name in cfg.DATASETS.TEST
|
59 |
+
]
|
60 |
+
res = cls.test(cfg, model, evaluators)
|
61 |
+
res = OrderedDict({k + "_TTA": v for k, v in res.items()})
|
62 |
+
return res
|
63 |
+
|
64 |
+
|
65 |
+
def setup(args):
|
66 |
+
"""
|
67 |
+
Create configs and perform basic setups.
|
68 |
+
"""
|
69 |
+
cfg = get_cfg()
|
70 |
+
cfg.merge_from_file(args.config_file)
|
71 |
+
cfg.merge_from_list(args.opts)
|
72 |
+
|
73 |
+
with open(args.json_annotation_train, 'r') as fp:
|
74 |
+
anno_file = json.load(fp)
|
75 |
+
|
76 |
+
cfg.MODEL.ROI_HEADS.NUM_CLASSES = len(anno_file["categories"])
|
77 |
+
del anno_file
|
78 |
+
|
79 |
+
cfg.DATASETS.TRAIN = (f"{args.dataset_name}-train",)
|
80 |
+
cfg.DATASETS.TEST = (f"{args.dataset_name}-val",)
|
81 |
+
|
82 |
+
cfg.freeze()
|
83 |
+
default_setup(cfg, args)
|
84 |
+
return cfg
|
85 |
+
|
86 |
+
|
87 |
+
def main(args):
|
88 |
+
cfg = setup(args)
|
89 |
+
|
90 |
+
if args.eval_only:
|
91 |
+
model = Trainer.build_model(cfg)
|
92 |
+
DetectionCheckpointer(model, save_dir=cfg.OUTPUT_DIR).resume_or_load(
|
93 |
+
cfg.MODEL.WEIGHTS, resume=args.resume
|
94 |
+
)
|
95 |
+
res = Trainer.test(cfg, model)
|
96 |
+
|
97 |
+
if cfg.TEST.AUG.ENABLED:
|
98 |
+
res.update(Trainer.test_with_TTA(cfg, model))
|
99 |
+
if comm.is_main_process():
|
100 |
+
verify_results(cfg, res)
|
101 |
+
|
102 |
+
# Save the evaluation results
|
103 |
+
pd.DataFrame(res).to_csv(f'{cfg.OUTPUT_DIR}/eval.csv')
|
104 |
+
return res
|
105 |
+
|
106 |
+
"""
|
107 |
+
If you'd like to do anything fancier than the standard training logic,
|
108 |
+
consider writing your own training loop (see plain_train_net.py) or
|
109 |
+
subclassing the trainer.
|
110 |
+
"""
|
111 |
+
trainer = Trainer(cfg)
|
112 |
+
trainer.resume_or_load(resume=args.resume)
|
113 |
+
if cfg.TEST.AUG.ENABLED:
|
114 |
+
trainer.register_hooks(
|
115 |
+
[hooks.EvalHook(0, lambda: trainer.test_with_TTA(cfg, trainer.model))]
|
116 |
+
)
|
117 |
+
return trainer.train()
|
118 |
+
|
119 |
+
|
120 |
+
if __name__ == "__main__":
|
121 |
+
parser = default_argument_parser()
|
122 |
+
|
123 |
+
# Extra Configurations for dataset names and paths
|
124 |
+
parser.add_argument("--dataset_name", default="", help="The Dataset Name")
|
125 |
+
parser.add_argument("--json_annotation_train", default="", metavar="FILE", help="The path to the training set JSON annotation")
|
126 |
+
parser.add_argument("--image_path_train", default="", metavar="FILE", help="The path to the training set image folder")
|
127 |
+
parser.add_argument("--json_annotation_val", default="", metavar="FILE", help="The path to the validation set JSON annotation")
|
128 |
+
parser.add_argument("--image_path_val", default="", metavar="FILE", help="The path to the validation set image folder")
|
129 |
+
|
130 |
+
args = parser.parse_args()
|
131 |
+
print("Command Line Args:", args)
|
132 |
+
|
133 |
+
# Register Datasets
|
134 |
+
dataset_name = args.dataset_name
|
135 |
+
register_coco_instances(f"{dataset_name}-train", {},
|
136 |
+
args.json_annotation_train,
|
137 |
+
args.image_path_train)
|
138 |
+
|
139 |
+
register_coco_instances(f"{dataset_name}-val", {},
|
140 |
+
args.json_annotation_val,
|
141 |
+
args.image_path_val)
|
142 |
+
|
143 |
+
launch(
|
144 |
+
main,
|
145 |
+
args.num_gpus,
|
146 |
+
num_machines=args.num_machines,
|
147 |
+
machine_rank=args.machine_rank,
|
148 |
+
dist_url=args.dist_url,
|
149 |
+
args=(args,),
|
150 |
+
)
|