Skip to content

Commit 6e01ea5

Browse files
zjZSTUzjZSTU
committed
docs(ssd): 添加训练结果和训练日志
1 parent 5f4c2aa commit 6e01ea5

File tree

7 files changed

+149
-12
lines changed

7 files changed

+149
-12
lines changed

‎docs/index.md‎

Lines changed: 9 additions & 6 deletions
Original file line numberDiff line numberDiff line change
@@ -16,9 +16,12 @@
1616
## SSD算法
1717

1818
* 引言
19-
* 基础网络
20-
* 先验框
21-
* 匹配策略
22-
* 损失函数
23-
* Hard Negative Mining
24-
* variance变量
19+
* 原理解析
20+
* 基础网络
21+
* 先验框
22+
* 匹配策略
23+
* 损失函数
24+
* `Hard Negative Mining`
25+
* `variance`变量
26+
* 训练
27+
* 训练日志

‎docs/ssd/imgs/loss.png‎

63 KB
Loading

‎docs/ssd/imgs/lr.png‎

29.7 KB
Loading

‎docs/ssd/imgs/metrics.png‎

146 KB
Loading

‎docs/ssd/train_params.md‎

Lines changed: 2 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,2 @@
1+
2+
# 训练参数

‎docs/ssd/train_res.md‎

Lines changed: 129 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,129 @@
1+
2+
# 训练日志
3+
4+
## 训练结果
5+
6+
![](./imgs/loss.png)
7+
8+
![](./imgs/lr.png)
9+
10+
![](./imgs/metrics.png)
11+
12+
## 训练日志
13+
14+
```
15+
$ python train.py --config-file configs/vgg_ssd300_voc0712.yaml
16+
2020-05-16 11:10:04,780 SSD INFO: Using 1 GPUs
17+
2020-05-16 11:10:04,780 SSD INFO: Namespace(config_file='configs/vgg_ssd300_voc0712.yaml', distributed=False, eval_step=2500, local_rank=0, log_step=10, num_gpus=1, opts=[], save_step=2500, skip_test=False, use_tensorboard=True)
18+
2020-05-16 11:10:04,780 SSD INFO: Loaded configuration file configs/vgg_ssd300_voc0712.yaml
19+
2020-05-16 11:10:04,780 SSD INFO:
20+
MODEL:
21+
NUM_CLASSES: 21
22+
INPUT:
23+
IMAGE_SIZE: 300
24+
DATASETS:
25+
TRAIN: ("voc_2007_trainval", "voc_2012_trainval")
26+
TEST: ("voc_2007_test", )
27+
SOLVER:
28+
MAX_ITER: 120000
29+
LR_STEPS: [80000, 100000]
30+
GAMMA: 0.1
31+
BATCH_SIZE: 32
32+
LR: 1e-3
33+
34+
OUTPUT_DIR: 'outputs/vgg_ssd300_voc0712'
35+
2020-05-16 11:10:04,781 SSD INFO: Running with config:
36+
DATASETS:
37+
TEST: ('voc_2007_test',)
38+
TRAIN: ('voc_2007_trainval', 'voc_2012_trainval')
39+
DATA_LOADER:
40+
NUM_WORKERS: 8
41+
PIN_MEMORY: True
42+
INPUT:
43+
IMAGE_SIZE: 300
44+
PIXEL_MEAN: [123, 117, 104]
45+
MODEL:
46+
BACKBONE:
47+
NAME: vgg
48+
OUT_CHANNELS: (512, 1024, 512, 256, 256, 256)
49+
PRETRAINED: True
50+
BOX_HEAD:
51+
NAME: SSDBoxHead
52+
PREDICTOR: SSDBoxPredictor
53+
CENTER_VARIANCE: 0.1
54+
DEVICE: cuda
55+
META_ARCHITECTURE: SSDDetector
56+
NEG_POS_RATIO: 3
57+
NUM_CLASSES: 21
58+
PRIORS:
59+
ASPECT_RATIOS: [[2], [2, 3], [2, 3], [2, 3], [2], [2]]
60+
BOXES_PER_LOCATION: [4, 6, 6, 6, 4, 4]
61+
CLIP: True
62+
FEATURE_MAPS: [38, 19, 10, 5, 3, 1]
63+
MAX_SIZES: [60, 111, 162, 213, 264, 315]
64+
MIN_SIZES: [30, 60, 111, 162, 213, 264]
65+
STRIDES: [8, 16, 32, 64, 100, 300]
66+
SIZE_VARIANCE: 0.2
67+
THRESHOLD: 0.5
68+
OUTPUT_DIR: outputs/vgg_ssd300_voc0712
69+
SOLVER:
70+
BATCH_SIZE: 32
71+
GAMMA: 0.1
72+
LR: 0.001
73+
LR_STEPS: [80000, 100000]
74+
MAX_ITER: 120000
75+
MOMENTUM: 0.9
76+
WARMUP_FACTOR: 0.3333333333333333
77+
WARMUP_ITERS: 500
78+
WEIGHT_DECAY: 0.0005
79+
TEST:
80+
BATCH_SIZE: 10
81+
CONFIDENCE_THRESHOLD: 0.01
82+
MAX_PER_CLASS: -1
83+
MAX_PER_IMAGE: 100
84+
NMS_THRESHOLD: 0.45
85+
2020-05-16 11:10:14,710 SSD.trainer INFO: No checkpoint found.
86+
2020-05-16 11:10:14,970 SSD.trainer INFO: Start training ...
87+
2020-05-16 11:10:43,388 SSD.trainer INFO: iter: 000010, lr: 0.00035, total_loss: 20.884 (20.884), reg_loss: 3.032 (3.032), cls_loss: 17.852 (17.852), time: 2.668 (2.668), eta: 3 days, 16:54:42, mem: 8288M
88+
2020-05-16 11:10:49,240 SSD.trainer INFO: iter: 000020, lr: 0.00036, total_loss: 15.886 (18.385), reg_loss: 2.946 (2.989), cls_loss: 12.940 (15.396), time: 0.585 (1.626), eta: 2 days, 6:12:16, mem: 8288M
89+
2020-05-16 11:10:55,108 SSD.trainer INFO: iter: 000030, lr: 0.00037, total_loss: 15.040 (17.270), reg_loss: 2.730 (2.903), cls_loss: 12.310 (14.367), time: 0.587 (1.280), eta: 1 day, 18:39:04, mem: 8288M
90+
2020-05-16 11:11:00,976 SSD.trainer INFO: iter: 000040, lr: 0.00039, total_loss: 14.594 (16.601), reg_loss: 2.765 (2.868), cls_loss: 11.829 (13.733), time: 0.587 (1.107), eta: 1 day, 12:52:26, mem: 8288M
91+
2020-05-16 11:11:06,836 SSD.trainer INFO: iter: 000050, lr: 0.00040, total_loss: 13.840 (16.049), reg_loss: 2.842 (2.863), cls_loss: 10.998 (13.186), time: 0.586 (1.002), eta: 1 day, 9:24:06, mem: 8288M
92+
2020-05-16 11:11:12,690 SSD.trainer INFO: iter: 000060, lr: 0.00041, total_loss: 11.895 (15.357), reg_loss: 2.800 (2.853)
93+
。。。
94+
。。。
95+
2020-05-17 07:20:21,323 SSD.trainer INFO: iter: 119940, lr: 0.00001, total_loss: 1.961 (2.557), reg_loss: 0.593 (0.754), cls_loss: 1.368 (1.803), time: 0.589 (0.605), eta: 0:00:36, mem: 8288M
96+
2020-05-17 07:20:27,208 SSD.trainer INFO: iter: 119950, lr: 0.00001, total_loss: 1.739 (2.557), reg_loss: 0.511 (0.754), cls_loss: 1.227 (1.803), time: 0.588 (0.605), eta: 0:00:30, mem: 8288M
97+
2020-05-17 07:20:33,106 SSD.trainer INFO: iter: 119960, lr: 0.00001, total_loss: 1.777 (2.557), reg_loss: 0.488 (0.753), cls_loss: 1.289 (1.803), time: 0.590 (0.605), eta: 0:00:24, mem: 8288M
98+
2020-05-17 07:20:38,987 SSD.trainer INFO: iter: 119970, lr: 0.00001, total_loss: 1.709 (2.557), reg_loss: 0.463 (0.753), cls_loss: 1.246 (1.803), time: 0.588 (0.605), eta: 0:00:18, mem: 8288M
99+
2020-05-17 07:20:44,867 SSD.trainer INFO: iter: 119980, lr: 0.00001, total_loss: 1.843 (2.557), reg_loss: 0.528 (0.753), cls_loss: 1.315 (1.803), time: 0.588 (0.605), eta: 0:00:12, mem: 8288M
100+
2020-05-17 07:20:50,752 SSD.trainer INFO: iter: 119990, lr: 0.00001, total_loss: 1.918 (2.557), reg_loss: 0.546 (0.753), cls_loss: 1.371 (1.803), time: 0.588 (0.605), eta: 0:00:06, mem: 8288M
101+
2020-05-17 07:20:56,633 SSD.trainer INFO: iter: 120000, lr: 0.00001, total_loss: 1.885 (2.557), reg_loss: 0.529 (0.753), cls_loss: 1.356 (1.803), time: 0.588 (0.605), eta: 0:00:00, mem: 8288M
102+
2020-05-17 07:20:57,016 SSD.trainer INFO: Saving checkpoint to outputs/vgg_ssd300_voc0712/model_120000.pth
103+
2020-05-17 07:20:57,186 SSD.trainer INFO: Saving checkpoint to outputs/vgg_ssd300_voc0712/model_final.pth
104+
2020-05-17 07:20:57,302 SSD.trainer INFO: Total training time: 20:10:40 (0.6053 s / it)
105+
2020-05-17 07:20:57,314 SSD INFO: Start evaluating...
106+
2020-05-17 07:20:57,316 SSD.inference INFO: Evaluating voc_2007_test dataset(4952 images):
107+
100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 496/496 [00:40<00:00, 12.34it/s]
108+
2020-05-17 07:21:39,792 SSD.inference INFO: mAP: 0.7740
109+
aeroplane : 0.8113
110+
bicycle : 0.8322
111+
bird : 0.7476
112+
boat : 0.7160
113+
bottle : 0.5331
114+
bus : 0.8619
115+
car : 0.8669
116+
cat : 0.8781
117+
chair : 0.6243
118+
cow : 0.8297
119+
diningtable : 0.7607
120+
dog : 0.8393
121+
horse : 0.8634
122+
motorbike : 0.8420
123+
person : 0.7972
124+
pottedplant : 0.5075
125+
sheep : 0.7754
126+
sofa : 0.7775
127+
train : 0.8548
128+
tvmonitor : 0.7622
129+
```

‎mkdocs.yml‎

Lines changed: 9 additions & 6 deletions
Original file line numberDiff line numberDiff line change
@@ -66,9 +66,12 @@ nav:
6666
- 检查点读写: util/checkpoint.md
6767
- SSD算法:
6868
- 引言: ssd/index.md
69-
- 基础网络: ssd/base_network.md
70-
- 先验框: ssd/prior_boxes.md
71-
- 匹配策略: ssd/matching_strategy.md
72-
- 损失函数: ssd/loss_function.md
73-
- 'Hard Negative Mining': ssd/hard_negative_mining.md
74-
- variance变量: ssd/variance.md
69+
- 原理解析:
70+
- 基础网络: ssd/base_network.md
71+
- 先验框: ssd/prior_boxes.md
72+
- 匹配策略: ssd/matching_strategy.md
73+
- 损失函数: ssd/loss_function.md
74+
- 'Hard Negative Mining': ssd/hard_negative_mining.md
75+
- variance变量: ssd/variance.md
76+
- 训练:
77+
- 训练日志: ssd/train_res.md

0 commit comments

Comments
 (0)