PulseFocusPlatform/static/configs/anchor_free/pafnet_10x_coco.yml

171 lines
3.5 KiB
YAML

architecture: TTFNet
use_gpu: true
max_iters: 150000
log_smooth_window: 20
save_dir: output
snapshot_iter: 10000
metric: COCO
pretrain_weights: https://paddle-imagenet-models-name.bj.bcebos.com/ResNet50_vd_ssld_pretrained.tar
weights: output/pafnet_10x_coco/model_final
num_classes: 80
use_ema: true
ema_decay: 0.9998
TTFNet:
backbone: ResNet
ttf_head: TTFHead
ResNet:
norm_type: sync_bn
freeze_at: 0
freeze_norm: false
norm_decay: 0.
depth: 50
feature_maps: [2, 3, 4, 5]
variant: d
dcn_v2_stages: [3, 4, 5]
TTFHead:
head_conv: 128
wh_conv: 64
hm_head_conv_num: 2
wh_head_conv_num: 2
wh_offset_base: 16
wh_loss: GiouLoss
dcn_head: True
GiouLoss:
loss_weight: 5.
do_average: false
use_class_weight: false
LearningRate:
base_lr: 0.015
schedulers:
- !PiecewiseDecay
gamma: 0.1
milestones:
- 112500
- 137500
- !LinearWarmup
start_factor: 0.2
steps: 500
OptimizerBuilder:
optimizer:
momentum: 0.9
type: Momentum
regularizer:
factor: 0.0004
type: L2
TrainReader:
inputs_def:
fields: ['image', 'ttf_heatmap', 'ttf_box_target', 'ttf_reg_weight']
dataset:
!COCODataSet
image_dir: train2017
anno_path: annotations/instances_train2017.json
dataset_dir: dataset/coco
with_background: false
sample_transforms:
- !DecodeImage
to_rgb: true
with_cutmix: True
- !CutmixImage
alpha: 1.5
beta: 1.5
- !ColorDistort
hue: [-18., 18., 0.5]
saturation: [0.5, 1.5, 0.5]
contrast: [0.5, 1.5, 0.5]
brightness: [-32., 32., 0.5]
random_apply: False
hsv_format: True
random_channel: True
- !RandomExpand
ratio: 4
prob: 0.5
fill_value: [123.675, 116.28, 103.53]
- !RandomCrop
aspect_ratio: NULL
cover_all_box: True
- !RandomFlipImage
prob: 0.5
batch_transforms:
- !RandomShape
sizes: [416, 448, 480, 512, 544, 576, 608, 640, 672]
random_inter: True
resize_box: True
- !NormalizeImage
is_channel_first: false
is_scale: false
mean: [123.675, 116.28, 103.53]
std: [58.395, 57.12, 57.375]
- !Permute
to_bgr: false
channel_first: true
- !Gt2TTFTarget
num_classes: 80
down_ratio: 4
- !PadBatch
pad_to_stride: 32
batch_size: 12
shuffle: true
worker_num: 8
bufsize: 2
use_process: false
cutmix_epoch: 100
EvalReader:
inputs_def:
image_shape: [3, 512, 512]
fields: ['image', 'im_id', 'scale_factor']
dataset:
!COCODataSet
image_dir: val2017
anno_path: annotations/instances_val2017.json
dataset_dir: dataset/coco
with_background: false
sample_transforms:
- !DecodeImage
to_rgb: True
- !Resize
target_dim: 512
- !NormalizeImage
mean: [123.675, 116.28, 103.53]
std: [58.395, 57.12, 57.375]
is_scale: false
is_channel_first: false
- !Permute
to_bgr: false
channel_first: True
batch_size: 1
drop_empty: false
worker_num: 8
bufsize: 16
TestReader:
inputs_def:
image_shape: [3, 512, 512]
fields: ['image', 'im_id', 'scale_factor']
dataset:
!ImageFolder
anno_path: annotations/instances_val2017.json
with_background: false
sample_transforms:
- !DecodeImage
to_rgb: True
- !Resize
interp: 1
target_dim: 512
- !NormalizeImage
mean: [123.675, 116.28, 103.53]
std: [58.395, 57.12, 57.375]
is_scale: false
is_channel_first: false
- !Permute
to_bgr: false
channel_first: True
batch_size: 1