Skip to content

Supports for DETR training mode in process. #3963

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
ZwwWayne merged 13 commits into from
Nov 26, 2020
Merged

Supports for DETR training mode in process. #3963

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
13 commits
Select commit Hold shift + click to select a range
e16dc87
add input_img_shape in img_meta, and supports training mode toy
v-qjqs Oct 20, 2020
5e615ac
toy commit
v-qjqs Oct 27, 2020
57f38e8
toy commit
v-qjqs Nov 2, 2020
1a4b7c1
re-organize and support for detr training mode
v-qjqs Nov 12, 2020
5a9659f
add unit test for detr training, testing, and transformer_head
v-qjqs Nov 25, 2020
f4bbc64
add more unit test for transformer
v-qjqs Nov 25, 2020
90fe254
reformat
v-qjqs Nov 25, 2020
6dcc25b
remove unnecessary
v-qjqs Nov 25, 2020
07317cc
replace RandomCropWarpper with re-implemented RandomCrop, and merge f…
v-qjqs Nov 25, 2020
5c3fecc
use small transformer model for unit test
v-qjqs Nov 25, 2020
85ce11d
reformat docstring
v-qjqs Nov 25, 2020
9f0607a
reformat
v-qjqs Nov 26, 2020
8c2a9aa
reformat
v-qjqs Nov 26, 2020
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
67 changes: 50 additions & 17 deletions configs/detr/detr_r50_8x4_1x_coco.py → configs/detr/detr_r50_8x2_300e_coco.py
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,6 +1,5 @@
_base_ = [
'../_base_/datasets/coco_detection.py',
'../_base_/schedules/schedule_1x.py', '../_base_/default_runtime.py'
'../_base_/datasets/coco_detection.py', '../_base_/default_runtime.py'
]
model = dict(
type='DETR',
Expand All @@ -11,7 +10,7 @@
num_stages=4,
out_indices=(3, ),
frozen_stages=1,
norm_cfg=dict(type='BN', requires_grad=True),
norm_cfg=dict(type='BN', requires_grad=False),
norm_eval=True,
style='pytorch'),
bbox_head=dict(
Expand Down Expand Up @@ -45,8 +44,8 @@
# training and testing settings
train_cfg = dict(
assigner=dict(
type='HungarianMatcher', cls_weight=1., bbox_weight=5., iou_weight=2.),
pos_weight=-1)
type='HungarianAssigner', cls_weight=1., bbox_weight=5.,
iou_weight=2.))
test_cfg = dict(max_per_img=100)
img_norm_cfg = dict(
mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_rgb=True)
Expand All @@ -55,14 +54,39 @@
train_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='LoadAnnotations', with_bbox=True),
dict(
type='Resize',
img_scale=[(480, 1333), (512, 1333), (544, 1333), (576, 1333),
(608, 1333), (640, 1333), (672, 1333), (704, 1333),
(736, 1333), (768, 1333), (800, 1333)],
multiscale_mode='value',
keep_ratio=True),
dict(type='RandomFlip', flip_ratio=0.5),
dict(
type='AutoAugment',
policies=[[
dict(
type='Resize',
img_scale=[(480, 1333), (512, 1333), (544, 1333), (576, 1333),
(608, 1333), (640, 1333), (672, 1333), (704, 1333),
(736, 1333), (768, 1333), (800, 1333)],
multiscale_mode='value',
keep_ratio=True)
],
[
dict(
type='Resize',
img_scale=[(400, 1333), (500, 1333), (600, 1333)],
multiscale_mode='value',
keep_ratio=True),
dict(
type='RandomCrop',
crop_type='absolute_range',
crop_size=(384, 600),
allow_negative_crop=True),
dict(
type='Resize',
img_scale=[(480, 1333), (512, 1333), (544, 1333),
(576, 1333), (608, 1333), (640, 1333),
(672, 1333), (704, 1333), (736, 1333),
(768, 1333), (800, 1333)],
multiscale_mode='value',
override=True,
keep_ratio=True)
]]),
dict(type='Normalize', **img_norm_cfg),
dict(type='Pad', size_divisor=1),
dict(type='DefaultFormatBundle'),
Expand All @@ -87,11 +111,20 @@
])
]
data = dict(
samples_per_gpu=4,
workers_per_gpu=4,
samples_per_gpu=2,
workers_per_gpu=2,
train=dict(pipeline=train_pipeline),
val=dict(pipeline=test_pipeline),
test=dict(pipeline=test_pipeline))
# TODO optimizer

# TODO learning policy
# optimizer
optimizer = dict(
type='AdamW',
lr=0.0001,
weight_decay=0.0001,
paramwise_cfg=dict(
custom_keys={'backbone': dict(lr_mult=0.1, decay_mult=1.0)}))
optimizer_config = dict(grad_clip=dict(max_norm=0.1, norm_type=2))
# learning policy
lr_config = dict(policy='step', step=[200])
total_epochs = 300
dist_params = dict(_delete_=True, backend='nccl', port=29504)
18 changes: 15 additions & 3 deletions mmdet/datasets/pipelines/transforms.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -52,14 +52,19 @@ class Resize(object):
backend (str): Image resize backend, choices are 'cv2' and 'pillow'.
These two backends generates slightly different results. Defaults
to 'cv2'.
override (bool): Whether to override `scale` and `scale_factor` so as
to call resize twice. Default False. If True, after the first
resizing, the existed `scale` and `scale_factor` will be ignored
so the second resizing can be allowed.
"""

def __init__(self,
img_scale=None,
multiscale_mode='range',
ratio_range=None,
keep_ratio=True,
backend='cv2'):
backend='cv2',
override=False):
if img_scale is None:
self.img_scale = None
else:
Expand All @@ -80,6 +85,7 @@ def __init__(self,
self.multiscale_mode = multiscale_mode
self.ratio_range = ratio_range
self.keep_ratio = keep_ratio
self.override = override

@staticmethod
def random_select(img_scales):
Expand Down Expand Up @@ -275,8 +281,14 @@ def __call__(self, results):
else:
self._random_scale(results)
else:
assert 'scale_factor' not in results, (
'scale and scale_factor cannot be both set.')
if not self.override:
assert 'scale_factor' not in results, (
'scale and scale_factor cannot be both set.')
else:
results.pop('scale')
if 'scale_factor' in results:
results.pop('scale_factor')
self._random_scale(results)

self._resize_img(results)
self._resize_bboxes(results)
Expand Down
Loading