support dior-r

.gitignore

@@ -146,7 +146,6 @@ __pycache__/
 *double_head*
 *.DS_Store*
 *redet*
-*cfgs.py
 
 tools/*/test_dota/*
 tools/*/test_hrsc/*

configs/DIOR-R/csl/cfgs_res50_dior_r_csl_v1.py

@@ -0,0 +1,62 @@
+# -*- coding: utf-8 -*-
+from __future__ import division, print_function, absolute_import
+
+import numpy as np
+
+from alpharotate.utils.pretrain_zoo import PretrainModelZoo
+from configs._base_.models.retinanet_r50_fpn import *
+from configs._base_.datasets.dota_detection import *
+from configs._base_.schedules.schedule_1x import *
+
+# schedule
+BATCH_SIZE = 1
+GPU_GROUP = "0,1"
+NUM_GPU = len(GPU_GROUP.strip().split(','))
+SAVE_WEIGHTS_INTE = 11725 * 2
+DECAY_EPOCH = [8, 11, 20]
+MAX_EPOCH = 12
+WARM_EPOCH = 1 / 16.
+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
+DATASET_NAME = 'DIOR-R'
+CLASS_NUM = 20
+
+# 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')
+
+# bbox head
+NUM_SUBNET_CONV = 4
+LEVEL = ['P3', 'P4', 'P5', 'P6', 'P7']
+BASE_ANCHOR_SIZE_LIST = [32, 64, 128, 256, 512]
+ANCHOR_STRIDE = [8, 16, 32, 64, 128]
+ANCHOR_SCALES = [2 ** 0, 2 ** (1.0 / 3.0), 2 ** (2.0 / 3.0)]
+ANCHOR_RATIOS = [1, 1 / 2, 2.]
+ANGLE_RANGE = 180
+
+# loss
+CLS_WEIGHT = 1.0
+REG_WEIGHT = 1.0
+ANGLE_WEIGHT = 0.5
+REG_LOSS_MODE = None
+
+# CSL
+LABEL_TYPE = 0  # {0: gaussian_label, 1: rectangular_label, 2: pulse_label, 3: triangle_label}
+RADIUS = 6
+OMEGA = 1
+
+VERSION = 'RetinaNet_DIOR_R_CSL_2x_20200912'
+
+"""
+gaussian label, omega=1, r=6
+FLOPs: 761359510;    Trainable params: 36287541
+
+
+"""
+
+

configs/cfgs.py

@@ -0,0 +1,69 @@
+# -*- coding: utf-8 -*-
+from __future__ import division, print_function, absolute_import
+
+import numpy as np
+
+from alpharotate.utils.pretrain_zoo import PretrainModelZoo
+from configs._base_.models.retinanet_r50_fpn import *
+from configs._base_.datasets.dota_detection import *
+from configs._base_.schedules.schedule_1x import *
+
+# schedule
+BATCH_SIZE = 1
+GPU_GROUP = "0"
+NUM_GPU = len(GPU_GROUP.strip().split(','))
+LR = 1e-3
+SAVE_WEIGHTS_INTE = 27000
+DECAY_STEP = np.array(DECAY_EPOCH, np.int32) * SAVE_WEIGHTS_INTE
+MAX_ITERATION = SAVE_WEIGHTS_INTE * MAX_EPOCH
+WARM_EPOCH = 1. / 8.
+WARM_SETP = int(WARM_EPOCH * SAVE_WEIGHTS_INTE)
+
+# dataset
+
+# model
+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 = 2.0
+REG_LOSS_MODE = 3  # KLD loss
+
+KL_TAU = 2.0
+KL_FUNC = 0   # 0: sqrt  1: log
+
+VERSION = 'RetinaNet_DOTA_KL_1x_20210201'
+
+"""
+RetinaNet-H + kl (fix bug) + sqrt + tau=2
+FLOPs: 484911761;    Trainable params: 33002916
+This is your result for task 1:
+
+mAP: 0.7128325571761713
+ap of each class:
+plane:0.884289204525325,
+baseball-diamond:0.7653565915743398,
+bridge:0.440047662898936,
+ground-track-field:0.698238996872059,
+small-vehicle:0.7444580421686285,
+large-vehicle:0.7248184249702364,
+ship:0.843025375274411,
+tennis-court:0.8939539261877734,
+basketball-court:0.806571747424402,
+storage-tank:0.800303800899483,
+soccer-ball-field:0.5787146175790521,
+roundabout:0.6505316568373755,
+harbor:0.6554363687620437,
+swimming-pool:0.6686981948609415,
+helicopter:0.5380437468075643
+
+The submitted information is :
+
+Description: RetinaNet_DOTA_KL_1x_20210201_45.9w
+Username: SJTU-Det
+Institute: SJTU
+Emailadress: yangxue-2019-sjtu@sjtu.edu.cn
+TeamMembers: yangxue
+"""

tools/csl/train.py

@@ -20,7 +20,7 @@
 from configs import cfgs
 from alpharotate.libs.models.detectors.csl import build_whole_network
 from alpharotate.libs.utils.coordinate_convert import backward_convert, get_horizen_minAreaRectangle
-from alpharotate.utils import angle_smooth_label
+from alpharotate.utils.smooth_label import angle_smooth_label
 from alpharotate.libs.utils.coordinate_convert import coordinate_present_convert
 from alpharotate.utils.pretrain_zoo import PretrainModelZoo