mmdeploy
mmdeploy copied to clipboard
Published
20 hours ago •
open-mmlab
open-mmlab
[Bug] Couldn't get the performance after i deploy the Dino model which traned by my own dataset.
Checklist
- [X] I have searched related issues but cannot get the expected help.
- [X] 2. I have read the FAQ documentation but cannot get the expected help.
- [X] 3. The bug has not been fixed in the latest version.
Describe the bug
I trained the dino model with my own dataset, and deploy in the jetson Xavier NX, but i can't get the performance when i test the model.
he output result output_tensorrt.jpg is not aligned with output_pytorch.jpg,
thetest.pyoutput:
and i had used the balloon dataset to train the dino, it had the same problem:
pytorch:
tensorrt:
the test.py output:
And i had trained faster r-cnn with my own dataset, it had the performance:
I had trained dino with the official coco dataset, it still had the performance:
So maybe the dino model had some bug that can't deploy correctly when it trained with other dataset?
Reproduction
script:
python tools/deploy.py configs/mmdet/detection/detection_tensorrt_dynamic-320x320-1344x1344.py dino_balloon/dino-4scale_r50_8xb2-12e_coco.py dino_balloon/epoch_12.pth 351678851_e2aeebdafd_b.jpg --work-dir work_dir/dino_balloon --show --device cuda:0 --dump-info
And i had not change the config:
auto_scale_lr = dict(base_batch_size=4)
backend_args = None
data_root = '/home/jetson/mmdeploy11/weed/'
dataset_type = 'CocoDataset'
default_hooks = dict(
checkpoint=dict(interval=1, type='CheckpointHook'),
logger=dict(interval=50, type='LoggerHook'),
param_scheduler=dict(type='ParamSchedulerHook'),
sampler_seed=dict(type='DistSamplerSeedHook'),
timer=dict(type='IterTimerHook'),
visualization=dict(type='DetVisualizationHook'))
default_scope = 'mmdet'
env_cfg = dict(
cudnn_benchmark=False,
dist_cfg=dict(backend='nccl'),
mp_cfg=dict(mp_start_method='fork', opencv_num_threads=0))
launcher = 'none'
load_from = '/home/jetson/mmdeploy/dino123/epoch_36.pth'
log_level = 'INFO'
log_processor = dict(by_epoch=True, type='LogProcessor', window_size=50)
max_epochs = 36
model = dict(
as_two_stage=True,
backbone=dict(
depth=50,
frozen_stages=1,
init_cfg=dict(checkpoint='torchvision://resnet50', type='Pretrained'),
norm_cfg=dict(requires_grad=False, type='BN'),
norm_eval=True,
num_stages=4,
out_indices=(
1,
2,
3,
),
style='pytorch',
type='ResNet'),
bbox_head=dict(
loss_bbox=dict(loss_weight=5.0, type='L1Loss'),
loss_cls=dict(
alpha=0.25,
gamma=2.0,
loss_weight=1.0,
type='FocalLoss',
use_sigmoid=True),
loss_iou=dict(loss_weight=2.0, type='GIoULoss'),
num_classes=8,
sync_cls_avg_factor=True,
type='DINOHead'),
data_preprocessor=dict(
bgr_to_rgb=True,
mean=[
123.675,
116.28,
103.53,
],
pad_size_divisor=1,
std=[
58.395,
57.12,
57.375,
],
type='DetDataPreprocessor'),
decoder=dict(
layer_cfg=dict(
cross_attn_cfg=dict(dropout=0.0, embed_dims=256, num_levels=4),
ffn_cfg=dict(
embed_dims=256, feedforward_channels=2048, ffn_drop=0.0),
self_attn_cfg=dict(dropout=0.0, embed_dims=256, num_heads=8)),
num_layers=6,
post_norm_cfg=None,
return_intermediate=True),
dn_cfg=dict(
box_noise_scale=1.0,
group_cfg=dict(dynamic=True, num_dn_queries=100, num_groups=None),
label_noise_scale=0.5),
encoder=dict(
layer_cfg=dict(
ffn_cfg=dict(
embed_dims=256, feedforward_channels=2048, ffn_drop=0.0),
self_attn_cfg=dict(dropout=0.0, embed_dims=256, num_levels=4)),
num_layers=6),
neck=dict(
act_cfg=None,
in_channels=[
512,
1024,
2048,
],
kernel_size=1,
norm_cfg=dict(num_groups=32, type='GN'),
num_outs=4,
out_channels=256,
type='ChannelMapper'),
num_queries=900,
positional_encoding=dict(
normalize=True, num_feats=128, offset=0.0, temperature=20),
test_cfg=dict(max_per_img=300),
train_cfg=dict(
assigner=dict(
match_costs=[
dict(type='FocalLossCost', weight=2.0),
dict(box_format='xywh', type='BBoxL1Cost', weight=5.0),
dict(iou_mode='giou', type='IoUCost', weight=2.0),
],
type='HungarianAssigner')),
type='DINO',
with_box_refine=True)
optim_wrapper = dict(
clip_grad=dict(max_norm=0.1, norm_type=2),
optimizer=dict(lr=0.0001, type='AdamW', weight_decay=0.0001),
paramwise_cfg=dict(custom_keys=dict(backbone=dict(lr_mult=0.1))),
type='OptimWrapper')
param_scheduler = [
dict(
begin=0,
by_epoch=True,
end=36,
gamma=0.1,
milestones=[
30,
],
type='MultiStepLR'),
]
resume = False
test_cfg = dict(type='TestLoop')
test_dataloader = dict(
batch_size=2,
dataset=dict(
ann_file=
'/home/jetson/mmdeploy11/weed/annotations/test.json',
data_prefix=dict(img='test/'),
data_root='/home/jetson/mmdeploy11/weed/',
pipeline=[
dict(backend_args=None, type='LoadImageFromFile'),
dict(keep_ratio=True, scale=(
680,
680,
), type='Resize'),
dict(type='LoadAnnotations', with_bbox=True),
dict(
meta_keys=(
'img_id',
'img_path',
'ori_shape',
'img_shape',
'scale_factor',
),
type='PackDetInputs'),
],
test_mode=True,
type='CocoDataset'),
drop_last=False,
num_workers=2,
persistent_workers=True,
sampler=dict(shuffle=False, type='DefaultSampler'))
test_evaluator = dict(
ann_file=
'/home/jetson/mmdeploy11/weed/annotations/test.json',
format_only=True,
metric='bbox',
outfile_prefix='./work_dirs/coco_detection/test',
type='CocoMetric')
test_pipeline = [
dict(backend_args=None, type='LoadImageFromFile'),
dict(keep_ratio=True, scale=(
680,
680,
), type='Resize'),
dict(type='LoadAnnotations', with_bbox=True),
dict(
meta_keys=(
'img_id',
'img_path',
'ori_shape',
'img_shape',
'scale_factor',
),
type='PackDetInputs'),
]
train_cfg = dict(max_epochs=36, type='EpochBasedTrainLoop', val_interval=1)
train_dataloader = dict(
batch_sampler=dict(type='AspectRatioBatchSampler'),
batch_size=4,
dataset=dict(
ann_file='annotations/train.json',
backend_args=None,
data_prefix=dict(img='train/'),
data_root='/home/jetson/mmdeploy11/weed/',
filter_cfg=dict(filter_empty_gt=False, min_size=32),
pipeline=[
dict(backend_args=None, type='LoadImageFromFile'),
dict(type='LoadAnnotations', with_bbox=True),
dict(prob=0.5, type='RandomFlip'),
dict(
transforms=[
[
dict(
keep_ratio=True,
scales=[
(
480,
1333,
),
(
512,
1333,
),
(
544,
1333,
),
(
576,
1333,
),
(
608,
1333,
),
(
640,
1333,
),
(
672,
1333,
),
(
704,
1333,
),
(
736,
1333,
),
(
768,
1333,
),
(
800,
1333,
),
],
type='RandomChoiceResize'),
],
[
dict(
keep_ratio=True,
scales=[
(
400,
4200,
),
(
500,
4200,
),
(
600,
4200,
),
],
type='RandomChoiceResize'),
dict(
allow_negative_crop=True,
crop_size=(
384,
600,
),
crop_type='absolute_range',
type='RandomCrop'),
dict(
keep_ratio=True,
scales=[
(
480,
1333,
),
(
512,
1333,
),
(
544,
1333,
),
(
576,
1333,
),
(
608,
1333,
),
(
640,
1333,
),
(
672,
1333,
),
(
704,
1333,
),
(
736,
1333,
),
(
768,
1333,
),
(
800,
1333,
),
],
type='RandomChoiceResize'),
],
],
type='RandomChoice'),
dict(type='PackDetInputs'),
],
type='CocoDataset'),
num_workers=2,
persistent_workers=True,
sampler=dict(shuffle=True, type='DefaultSampler'))
train_pipeline = [
dict(backend_args=None, type='LoadImageFromFile'),
dict(type='LoadAnnotations', with_bbox=True),
dict(prob=0.5, type='RandomFlip'),
dict(
transforms=[
[
dict(
keep_ratio=True,
scales=[
(
480,
1333,
),
(
512,
1333,
),
(
544,
1333,
),
(
576,
1333,
),
(
608,
1333,
),
(
640,
1333,
),
(
672,
1333,
),
(
704,
1333,
),
(
736,
1333,
),
(
768,
1333,
),
(
800,
1333,
),
],
type='RandomChoiceResize'),
],
[
dict(
keep_ratio=True,
scales=[
(
400,
4200,
),
(
500,
4200,
),
(
600,
4200,
),
],
type='RandomChoiceResize'),
dict(
allow_negative_crop=True,
crop_size=(
384,
600,
),
crop_type='absolute_range',
type='RandomCrop'),
dict(
keep_ratio=True,
scales=[
(
480,
1333,
),
(
512,
1333,
),
(
544,
1333,
),
(
576,
1333,
),
(
608,
1333,
),
(
640,
1333,
),
(
672,
1333,
),
(
704,
1333,
),
(
736,
1333,
),
(
768,
1333,
),
(
800,
1333,
),
],
type='RandomChoiceResize'),
],
],
type='RandomChoice'),
dict(type='PackDetInputs'),
]
val_cfg = dict(type='ValLoop')
val_dataloader = dict(
batch_size=2,
dataset=dict(
ann_file='annotations/val.json',
backend_args=None,
data_prefix=dict(img='val/'),
data_root='/home/jetson/mmdeploy11/weed/',
pipeline=[
dict(backend_args=None, type='LoadImageFromFile'),
dict(keep_ratio=True, scale=(
680,
680,
), type='Resize'),
dict(type='LoadAnnotations', with_bbox=True),
dict(
meta_keys=(
'img_id',
'img_path',
'ori_shape',
'img_shape',
'scale_factor',
),
type='PackDetInputs'),
],
test_mode=True,
type='CocoDataset'),
drop_last=False,
num_workers=2,
persistent_workers=True,
sampler=dict(shuffle=False, type='DefaultSampler'))
val_evaluator = dict(
ann_file=
'/home/jetson/mmdeploy11/weed/annotations/val.json',
backend_args=None,
format_only=False,
metric='bbox',
type='CocoMetric')
vis_backends = [
dict(type='LocalVisBackend'),
]
visualizer = dict(
name='visualizer',
type='DetLocalVisualizer',
vis_backends=[
dict(type='LocalVisBackend'),
])
work_dir = 'work_dirs/dino-4scale_r50_8xb2-36e_coco/'
Environment
python:3.8
pytorch:1.13.1+cu11
mmdetion:dev-3.x
mmdeploy:dev-1.x
Error traceback
No response
