mirrors
/
mmyolo
mirror of https://github.com/open-mmlab/mmyolo.git
1
0
Fork 0
Code Issues Projects Releases Wiki Activity
mmyolo/configs/ppyoloe/ppyoloe_plus_m_fast_8xb8-80...

17 lines
638 B
Python
Raw Normal View History

[Feature] Support PPYOLOE inference (#145) * add ppyoloe backbone, neck * add ppyoloe test * add docstring * add ppyoloe m/l/x configfile * add ppyoloe_coco.py * rename config * add typehint * format repvgg * format code; add ut * add doc; use mmdet.DropBlock * add doc, format code, add non-plus config * format code * format code * format code * format code * format code * format code
2022-10-28 09:44:11 +08:00
_base_ = './ppyoloe_plus_s_fast_8xb8-80e_coco.py'
[Feature] Support PPYOLOE training (#259) * add ppyoloe backbone, neck * add ppyoloe test * add docstring * add ppyoloe m/l/x configfile * add ppyoloe_coco.py * rename config * add typehint * format code; add ut * add datapre * add datapre * add ppyoloe datapre * add ppyoloe datapre * add ppyoloe datapre * reproduce coco v0.1 * add ut * add ut, docstring * fix transforms bug * use mmdet dfloss * add non plus model config * add non plus model config * fix * add ut * produce coco v0.2 * fix config * fix config * fix eps and transforms bug * add ema * fix resize * fix transforms.py * fix transforms.py * fix transforms.py * old version * old version * old version * old version * old version * old version * fix stride loss error * add INTER_LANCZOS4 * fix crop bug * init commit * format code * format code * bgr transforms.py * add typehint and doc in transforms.py * 继承新版yolov6head写法，删除不必要的注释 * fix transforms var name bug * bbox decode use stridetensor insted of priors * add headmodule todo * add ppyoloe README.md * add ppyoloe README.md * Update tests/test_datasets/test_transforms/test_transforms.py Co-authored-by: Range King <RangeKingHZ@gmail.com> * Update tests/test_datasets/test_transforms/test_transforms.py Co-authored-by: Range King <RangeKingHZ@gmail.com> * save ckpt last 10 epochs * save_best ckpt * del ppyoloe collate * change name of ppyoloebatchrandomresize * add the reason for rewritten PPYOLOEDetDataPreprocessor * solve ppyoloerandomresize name error * rm PPYOLOERandomExpand * rm l1 loss * rm yolov6 loss_obj * add difference between yolov6 and ppyoloe * add reason for rewrite paramscheduler * change proj init way * fix error * rm proj_conv in pth * format code * add load_from * update * support fast training * add pretrained model url * update * add pretrained model url * fix error * add imagenet model convert and use init_cfg to init backbone * add plus model pretrain model * add ut * add ut * fix ut * fix withstride bug * cat in yolov5_collate * merge * fix typehint * update readme * add reason for gap * fix log in README.md * rollback yolov6 * change inherit * fix ut * fix ut Co-authored-by: Range King <RangeKingHZ@gmail.com> Co-authored-by: hha <1286304229@qq.com> Co-authored-by: huanghaian <huanghaian@sensetime.com>
2023-01-06 15:54:39 +08:00
# The pretrained model is geted and converted from official PPYOLOE.
# https://github.com/PaddlePaddle/PaddleDetection/blob/release/2.5/configs/ppyoloe/README.md
load_from = 'https://download.openmmlab.com/mmyolo/v0/ppyoloe/ppyoloe_pretrain/ppyoloe_plus_m_ojb365_pretrained-03206892.pth' # noqa
[Feature] Support PPYOLOE inference (#145) * add ppyoloe backbone, neck * add ppyoloe test * add docstring * add ppyoloe m/l/x configfile * add ppyoloe_coco.py * rename config * add typehint * format repvgg * format code; add ut * add doc; use mmdet.DropBlock * add doc, format code, add non-plus config * format code * format code * format code * format code * format code * format code
2022-10-28 09:44:11 +08:00
deepen_factor = 0.67
widen_factor = 0.75
model = dict(
backbone=dict(deepen_factor=deepen_factor, widen_factor=widen_factor),
neck=dict(
deepen_factor=deepen_factor,
widen_factor=widen_factor,
),
bbox_head=dict(head_module=dict(widen_factor=widen_factor)))