-
Notifications
You must be signed in to change notification settings - Fork 181
/
Copy pathcfgs_res50_dota_fcos_v1.py
71 lines (56 loc) · 1.78 KB
/
cfgs_res50_dota_fcos_v1.py
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
# -*- coding: utf-8 -*-
from __future__ import division, print_function, absolute_import
import numpy as np
from configs._base_.models.retinanet_r50_fpn import *
from configs._base_.datasets.dota_detection import *
from configs._base_.schedules.schedule_1x import *
from alpharotate.utils.pretrain_zoo import PretrainModelZoo
# schedule
BATCH_SIZE = 1
GPU_GROUP = "0"
NUM_GPU = len(GPU_GROUP.strip().split(','))
LR = 1e-3 * BATCH_SIZE * NUM_GPU
SAVE_WEIGHTS_INTE = 27000
DECAY_STEP = np.array(DECAY_EPOCH, np.int32) * SAVE_WEIGHTS_INTE
MAX_ITERATION = SAVE_WEIGHTS_INTE * MAX_EPOCH
WARM_SETP = int(WARM_EPOCH * SAVE_WEIGHTS_INTE)
# dataset
# model
# backbone
pretrain_zoo = PretrainModelZoo()
PRETRAINED_CKPT = pretrain_zoo.pretrain_weight_path(NET_NAME, ROOT_PATH)
TRAINED_CKPT = os.path.join(ROOT_PATH, 'output/trained_weights')
# loss
CLS_WEIGHT = 1.0
REG_WEIGHT = 1.0
CTR_WEIGHT = 1.0
REG_LOSS_MODE = None
VERSION = 'FCOS_DOTA_1x_20210611'
"""
FCOS
FLOPs: 468484112; Trainable params: 32090136
This is your result for task 1:
mAP: 0.6768883021004449
ap of each class:
plane:0.872774857044874,
baseball-diamond:0.7456605755847489,
bridge:0.45562517475651065,
ground-track-field:0.5894727115456821,
small-vehicle:0.7427962860138525,
large-vehicle:0.6294806602885703,
ship:0.7589526279294885,
tennis-court:0.8899116518355858,
basketball-court:0.7648991572877125,
storage-tank:0.8169922109873226,
soccer-ball-field:0.536122347311571,
roundabout:0.5802443666640051,
harbor:0.6179626999644399,
swimming-pool:0.6741887143799705,
helicopter:0.47824048991233775
The submitted information is :
Description: FCOS_DOTA_1x_20210611_35.1w_
Username: SJTU-Det
Institute: SJTU
Emailadress: [email protected]
TeamMembers: yangxue
"""