提交 890f7741 编写于 作者: G gaotingquan 提交者: cuicheng01

fix bs and unset update_freq to adapt to 8 gpus

上级 fc9c59c4
......@@ -14,7 +14,6 @@ Global:
image_shape: [3, 256, 256]
save_inference_dir: ./inference
use_dali: False
update_freq: 3 # for 4 gpus
# mixed precision training
AMP:
......@@ -85,7 +84,7 @@ DataLoader:
scales: [256, 160, 192, 224, 288, 320]
# first_bs: batch size for the first image resolution in the scales list
# divide_factor: to ensure the width and height dimensions can be devided by downsampling multiple
first_bs: 32
first_bs: 48
divided_factor: 32
is_training: True
loader:
......
......@@ -14,7 +14,6 @@ Global:
image_shape: [3, 256, 256]
save_inference_dir: ./inference
use_dali: False
update_freq: 3 # for 4 gpus
# mixed precision training
AMP:
......@@ -85,7 +84,7 @@ DataLoader:
scales: [256, 160, 192, 224, 288, 320]
# first_bs: batch size for the first image resolution in the scales list
# divide_factor: to ensure the width and height dimensions can be devided by downsampling multiple
first_bs: 32
first_bs: 48
divided_factor: 32
is_training: True
loader:
......
......@@ -14,7 +14,6 @@ Global:
image_shape: [3, 256, 256]
save_inference_dir: ./inference
use_dali: False
update_freq: 3 # for 4 gpus
# mixed precision training
AMP:
......@@ -85,7 +84,7 @@ DataLoader:
scales: [256, 160, 192, 224, 288, 320]
# first_bs: batch size for the first image resolution in the scales list
# divide_factor: to ensure the width and height dimensions can be devided by downsampling multiple
first_bs: 32
first_bs: 48
divided_factor: 32
is_training: True
loader:
......
......@@ -14,7 +14,6 @@ Global:
image_shape: [3, 256, 256]
save_inference_dir: ./inference
use_dali: False
update_freq: 3 # for 4 gpus
# mixed precision training
AMP:
......@@ -85,7 +84,7 @@ DataLoader:
scales: [256, 160, 192, 224, 288, 320]
# first_bs: batch size for the first image resolution in the scales list
# divide_factor: to ensure the width and height dimensions can be devided by downsampling multiple
first_bs: 32
first_bs: 48
divided_factor: 32
is_training: True
loader:
......
......@@ -14,7 +14,6 @@ Global:
image_shape: [3, 256, 256]
save_inference_dir: ./inference
use_dali: False
update_freq: 3 # for 4 gpus
# mixed precision training
AMP:
......@@ -85,7 +84,7 @@ DataLoader:
scales: [256, 160, 192, 224, 288, 320]
# first_bs: batch size for the first image resolution in the scales list
# divide_factor: to ensure the width and height dimensions can be devided by downsampling multiple
first_bs: 32
first_bs: 48
divided_factor: 32
is_training: True
loader:
......
......@@ -14,7 +14,6 @@ Global:
image_shape: [3, 256, 256]
save_inference_dir: ./inference
use_dali: False
update_freq: 3 # for 4 gpus
# mixed precision training
AMP:
......@@ -85,7 +84,7 @@ DataLoader:
scales: [256, 160, 192, 224, 288, 320]
# first_bs: batch size for the first image resolution in the scales list
# divide_factor: to ensure the width and height dimensions can be devided by downsampling multiple
first_bs: 32
first_bs: 48
divided_factor: 32
is_training: True
loader:
......
......@@ -14,7 +14,6 @@ Global:
image_shape: [3, 256, 256]
save_inference_dir: ./inference
use_dali: False
update_freq: 3 # for 4 gpus
# mixed precision training
AMP:
......@@ -52,7 +51,7 @@ Optimizer:
one_dim_param_no_weight_decay: True
lr:
name: Cosine
learning_rate: 0.002 # for total batch size 1020
learning_rate: 0.002 # for total batch size 1020 by referring to official
eta_min: 0.0002
warmup_epoch: 16 # 20000 iterations
warmup_start_lr: 1e-6
......@@ -103,7 +102,7 @@ DataLoader:
prob: 0.25
sampler:
name: DistributedBatchSampler
batch_size: 85
batch_size: 128
drop_last: False
shuffle: True
loader:
......
......@@ -14,7 +14,6 @@ Global:
image_shape: [3, 256, 256]
save_inference_dir: ./inference
use_dali: False
update_freq: 3 # for 4 gpus
# mixed precision training
AMP:
......@@ -52,7 +51,7 @@ Optimizer:
one_dim_param_no_weight_decay: True
lr:
name: Cosine
learning_rate: 0.002 # for total batch size 1020
learning_rate: 0.002 # for total batch size 1020 by referring to official
eta_min: 0.0002
warmup_epoch: 16 # 20000 iterations
warmup_start_lr: 1e-6
......@@ -103,7 +102,7 @@ DataLoader:
prob: 0.25
sampler:
name: DistributedBatchSampler
batch_size: 85
batch_size: 128
drop_last: False
shuffle: True
loader:
......
......@@ -14,7 +14,6 @@ Global:
image_shape: [3, 256, 256]
save_inference_dir: ./inference
use_dali: False
update_freq: 3 # for 4 gpus
# mixed precision training
AMP:
......@@ -52,7 +51,7 @@ Optimizer:
one_dim_param_no_weight_decay: True
lr:
name: Cosine
learning_rate: 0.002 # for total batch size 1020
learning_rate: 0.002 # for total batch size 1020 by referring to official
eta_min: 0.0002
warmup_epoch: 16 # 20000 iterations
warmup_start_lr: 1e-6
......@@ -103,7 +102,7 @@ DataLoader:
prob: 0.25
sampler:
name: DistributedBatchSampler
batch_size: 85
batch_size: 128
drop_last: False
shuffle: True
loader:
......
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册