search

open-mmlab / mmdetection

OpenMMLab Detection Toolbox and Benchmark
https://mmdetection.readthedocs.io
Apache License 2.0
29.62k stars 9.47k forks source link

Training of faster R-cnn with resnet50 dc5 on voc07 is very slow #5485

Closed inder9999 closed 3 years ago

inder9999 commented 3 years ago

Describe the bug Training of faster R-cnn with resnet50 dc5 on voc07 is very slow. Have obtained only 0.208 mAP for 12 epochs.

Reproduction

./tools/dist_train.sh configs/pascal_voc/faster_rcnn_r50_caffe_dc5_voc0712.py 3 --cfg-options data.samples_per_gpu=3 optimizer.lr=0.005 runner.max_epochs=15

Dataset pascal voc07

Environment

sys.platform: linux
Python: 3.7.9 (default, Aug 18 2020, 06:24:24) [GCC 5.4.0 20160609]
CUDA available: True
GPU 0,1,2: NVIDIA GeForce GTX 1080 Ti
CUDA_HOME: /usr/local/cuda
NVCC: Cuda compilation tools, release 9.0, V9.0.176
GCC: gcc (Ubuntu 6.5.0-2ubuntu1~16.04) 6.5.0 20181026
PyTorch: 1.8.0+cu111
PyTorch compiling details: PyTorch built with:
  - GCC 7.3
  - C++ Version: 201402
  - Intel(R) Math Kernel Library Version 2020.0.0 Product Build 20191122 for Intel(R) 64 architecture applications
  - Intel(R) MKL-DNN v1.7.0 (Git Hash 7aed236906b1f7a05c0917e5257a1af05e9ff683)
  - OpenMP 201511 (a.k.a. OpenMP 4.5)
  - NNPACK is enabled
  - CPU capability usage: AVX2
  - CUDA Runtime 11.1
  - NVCC architecture flags: -gencode;arch=compute_37,code=sm_37;-gencode;arch=compute_50,code=sm_50;-gencode;arch=compute_60,code=sm_60;-gencode;arch=compute_70,code=sm_70;-gencode;arch=compute_75,code=sm_75;-gencode;arch=compute_80,code=sm_80;-gencode;arch=compute_86,code=sm_86
  - CuDNN 8.0.5
  - Magma 2.5.2
  - Build settings: BLAS_INFO=mkl, BUILD_TYPE=Release, CUDA_VERSION=11.1, CUDNN_VERSION=8.0.5, CXX_COMPILER=/opt/rh/devtoolset-7/root/usr/bin/c++, CXX_FLAGS= -Wno-deprecated -fvisibility-inlines-hidden -DUSE_PTHREADPOOL -fopenmp -DNDEBUG -DUSE_KINETO -DUSE_FBGEMM -DUSE_QNNPACK -DUSE_PYTORCH_QNNPACK -DUSE_XNNPACK -O2 -fPIC -Wno-narrowing -Wall -Wextra -Werror=return-type -Wno-missing-field-initializers -Wno-type-limits -Wno-array-bounds -Wno-unknown-pragmas -Wno-sign-compare -Wno-unused-parameter -Wno-unused-variable -Wno-unused-function -Wno-unused-result -Wno-unused-local-typedefs -Wno-strict-overflow -Wno-strict-aliasing -Wno-error=deprecated-declarations -Wno-stringop-overflow -Wno-psabi -Wno-error=pedantic -Wno-error=redundant-decls -Wno-error=old-style-cast -fdiagnostics-color=always -faligned-new -Wno-unused-but-set-variable -Wno-maybe-uninitialized -fno-math-errno -fno-trapping-math -Werror=format -Wno-stringop-overflow, LAPACK_INFO=mkl, PERF_WITH_AVX=1, PERF_WITH_AVX2=1, PERF_WITH_AVX512=1, TORCH_VERSION=1.8.0, USE_CUDA=ON, USE_CUDNN=ON, USE_EXCEPTION_PTR=1, USE_GFLAGS=OFF, USE_GLOG=OFF, USE_MKL=ON, USE_MKLDNN=ON, USE_MPI=OFF, USE_NCCL=ON, USE_NNPACK=ON, USE_OPENMP=ON,

TorchVision: 0.9.0+cu111
OpenCV: 4.5.2
MMCV: 1.3.7
MMCV Compiler: GCC 7.3
MMCV CUDA Compiler: 11.1
MMDetection: 2.13.0+0e09f00

Configs

INFO:mmdet:Config:
norm_cfg = dict(type='BN', requires_grad=False)
model = dict(
    type='FasterRCNN',
    pretrained='open-mmlab://detectron2/resnet50_caffe',
    backbone=dict(
        type='ResNet',
        depth=50,
        num_stages=4,
        strides=(1, 2, 2, 1),
        dilations=(1, 1, 1, 2),
        out_indices=(3, ),
        frozen_stages=1,
        norm_cfg=dict(type='BN', requires_grad=False),
        norm_eval=True,
        style='caffe'),
    rpn_head=dict(
        type='RPNHead',
        in_channels=2048,
        feat_channels=2048,
        anchor_generator=dict(
            type='AnchorGenerator',
            scales=[2, 4, 8, 16, 32],
            ratios=[0.5, 1.0, 2.0],
            strides=[16]),
        bbox_coder=dict(
            type='DeltaXYWHBBoxCoder',
            target_means=[0.0, 0.0, 0.0, 0.0],
            target_stds=[1.0, 1.0, 1.0, 1.0]),
        loss_cls=dict(
            type='CrossEntropyLoss', use_sigmoid=True, loss_weight=1.0),
        loss_bbox=dict(type='L1Loss', loss_weight=1.0)),
    roi_head=dict(
        type='StandardRoIHead',
        bbox_roi_extractor=dict(
            type='SingleRoIExtractor',
            roi_layer=dict(type='RoIAlign', output_size=7, sampling_ratio=0),
            out_channels=2048,
            featmap_strides=[16]),
        bbox_head=dict(
            type='Shared2FCBBoxHead',
            in_channels=2048,
            fc_out_channels=1024,
            roi_feat_size=7,
            num_classes=20,
            bbox_coder=dict(
                type='DeltaXYWHBBoxCoder',
                target_means=[0.0, 0.0, 0.0, 0.0],
                target_stds=[0.1, 0.1, 0.2, 0.2]),
            reg_class_agnostic=False,
            loss_cls=dict(
                type='CrossEntropyLoss', use_sigmoid=False, loss_weight=1.0),
            loss_bbox=dict(type='L1Loss', loss_weight=1.0))),
    train_cfg=dict(
        rpn=dict(
            assigner=dict(
                type='MaxIoUAssigner',
                pos_iou_thr=0.7,
                neg_iou_thr=0.3,
                min_pos_iou=0.3,
                match_low_quality=True,
                ignore_iof_thr=-1),
            sampler=dict(
                type='RandomSampler',
                num=256,
                pos_fraction=0.5,
                neg_pos_ub=-1,
                add_gt_as_proposals=False),
            allowed_border=0,
            pos_weight=-1,
            debug=False),
        rpn_proposal=dict(
            nms_pre=12000,
            max_per_img=2000,
            nms=dict(type='nms', iou_threshold=0.7),
            min_bbox_size=0),
        rcnn=dict(
            assigner=dict(
                type='MaxIoUAssigner',
                pos_iou_thr=0.5,
                neg_iou_thr=0.5,
                min_pos_iou=0.5,
                match_low_quality=False,
                ignore_iof_thr=-1),
            sampler=dict(
                type='RandomSampler',
                num=512,
                pos_fraction=0.25,
                neg_pos_ub=-1,
                add_gt_as_proposals=True),
            pos_weight=-1,
            debug=False)),
    test_cfg=dict(
        rpn=dict(
            nms=dict(type='nms', iou_threshold=0.7),
            nms_pre=6000,
            max_per_img=1000,
            min_bbox_size=0),
        rcnn=dict(
            score_thr=0.05,
            nms=dict(type='nms', iou_threshold=0.5),
            max_per_img=100)))
dataset_type = 'VOCDataset'
data_root = 'data/VOCdevkit/'
img_norm_cfg = dict(
    mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_rgb=True)
train_pipeline = [
    dict(type='LoadImageFromFile'),
    dict(type='LoadAnnotations', with_bbox=True),
    dict(type='Resize', img_scale=(1000, 600), keep_ratio=True),
    dict(type='RandomFlip', flip_ratio=0.5),
    dict(
        type='Normalize',
        mean=[123.675, 116.28, 103.53],
        std=[58.395, 57.12, 57.375],
        to_rgb=True),
    dict(type='Pad', size_divisor=32),
    dict(type='DefaultFormatBundle'),
    dict(type='Collect', keys=['img', 'gt_bboxes', 'gt_labels'])
]
test_pipeline = [
    dict(type='LoadImageFromFile'),
    dict(
        type='MultiScaleFlipAug',
        img_scale=(1000, 600),
        flip=False,
        transforms=[
            dict(type='Resize', keep_ratio=True),
            dict(type='RandomFlip'),
            dict(
                type='Normalize',
                mean=[123.675, 116.28, 103.53],
                std=[58.395, 57.12, 57.375],
                to_rgb=True),
            dict(type='Pad', size_divisor=32),
            dict(type='ImageToTensor', keys=['img']),
            dict(type='Collect', keys=['img'])
        ])
]
data = dict(
    samples_per_gpu=3,
    workers_per_gpu=2,
    train=dict(
        type='RepeatDataset',
        times=3,
        dataset=dict(
            type='VOCDataset',
            ann_file=['data/VOCdevkit/VOC2007/ImageSets/Main/trainval.txt'],
            img_prefix=['data/VOCdevkit/VOC2007/'],
            pipeline=[
                dict(type='LoadImageFromFile'),
                dict(type='LoadAnnotations', with_bbox=True),
                dict(type='Resize', img_scale=(1000, 600), keep_ratio=True),
                dict(type='RandomFlip', flip_ratio=0.5),
                dict(
                    type='Normalize',
                    mean=[123.675, 116.28, 103.53],
                    std=[58.395, 57.12, 57.375],
                    to_rgb=True),
                dict(type='Pad', size_divisor=32),
                dict(type='DefaultFormatBundle'),
                dict(type='Collect', keys=['img', 'gt_bboxes', 'gt_labels'])
            ])),
    val=dict(
        type='VOCDataset',
        ann_file='data/VOCdevkit/VOC2007/ImageSets/Main/test.txt',
        img_prefix='data/VOCdevkit/VOC2007/',
        pipeline=[
            dict(type='LoadImageFromFile'),
            dict(
                type='MultiScaleFlipAug',
                img_scale=(1000, 600),
                flip=False,
                transforms=[
                    dict(type='Resize', keep_ratio=True),
                    dict(type='RandomFlip'),
                    dict(
                        type='Normalize',
                        mean=[123.675, 116.28, 103.53],
                        std=[58.395, 57.12, 57.375],
                        to_rgb=True),
                    dict(type='Pad', size_divisor=32),
                    dict(type='ImageToTensor', keys=['img']),
                    dict(type='Collect', keys=['img'])
                ])
        ]),
    test=dict(
        type='VOCDataset',
        ann_file='data/VOCdevkit/VOC2007/ImageSets/Main/test.txt',
        img_prefix='data/VOCdevkit/VOC2007/',
        pipeline=[
            dict(type='LoadImageFromFile'),
            dict(
                type='MultiScaleFlipAug',
                img_scale=(1000, 600),
                flip=False,
                transforms=[
                    dict(type='Resize', keep_ratio=True),
                    dict(type='RandomFlip'),
                    dict(
                        type='Normalize',
                        mean=[123.675, 116.28, 103.53],
                        std=[58.395, 57.12, 57.375],
                        to_rgb=True),
                    dict(type='Pad', size_divisor=32),
                    dict(type='ImageToTensor', keys=['img']),
                    dict(type='Collect', keys=['img'])
                ])
        ]))
evaluation = dict(interval=1, metric='mAP')
checkpoint_config = dict(interval=1)
log_config = dict(interval=50, hooks=[dict(type='TextLoggerHook')])
custom_hooks = [dict(type='NumClassCheckHook')]
dist_params = dict(backend='nccl')
log_level = 'INFO'
load_from = None
resume_from = None
workflow = [('train', 1)]
optimizer = dict(type='SGD', lr=0.005, momentum=0.9, weight_decay=0.0001)
optimizer_config = dict(grad_clip=None)
lr_config = dict(policy='step', step=[3])
runner = dict(type='EpochBasedRunner', max_epochs=15)
work_dir = './work_dirs/faster_rcnn_r50_caffe_dc5_voc0712'
gpu_ids = range(0, 3)

Bug fix If you have already identified the reason, you can provide the information here. If you are willing to create a PR to fix it, please also leave a comment here and that would be much appreciated!

AronLin commented 3 years ago

It seems like your learning rate is too small. You use Repeated Dataset and the repeat number is 3, you should set max_epochs=5 to train your model for 15 epochs.

inder9999 commented 3 years ago

Hi! @AronLin Thanks for your reply and time.

I have used the learning rate as described here (https://github.com/open-mmlab/mmdetection/blob/2a856efb6f79afe8daa5ac97bf1711ca41b5fbb0/docs/1_exist_data_model.md#:~:text=Important%3A%20The%20default,4%20imgs/gpu. ) So I think lr=0.005 should be fine.

I did the experiment with the same configurations as above on a single GPU with lr=0.001. I am able to get the mAP score of 65.2 when trained for max_epochs=5. The command that I used was :for

python tools/train.py configs/pascal_voc/faster_rcnn_r50_caffe_dc5_voc0712.py --work-dir work_dirs/res50dc5_voc_singlegpu --cfg-options data.samples_per_gpu=2 optimizer.lr=0.001 runner.max_epochs=5

Further I also did the same experiment with VGG16 backbone again on 3 GPUs and faced the same issue. I trained the model for max_epochs=30 and the mAP score was around 39. I think the problem occurs when I do training on Multiple GPUs. What do you think about it?

Thanks again for the help!!

AronLin commented 3 years ago

I trained faster_rcnn_r50_fpn_1x_voc0712 with 3 GPUs and got 80.9 mAP. I have not changed any settings but the number of GPUs. Here is the command:

GPUS=3 GPUS_PER_NODE=3 CPUS_PER_TASK=5  ./tools/slurm_train.sh openmmlab voc ./configs/pascal_voc/faster_rcnn_r50_fpn_1x_voc0712 --work-dir my_work_dir

You can try to train your model without setting --cfg-options and directly use the origin settings. Maybe setting lr=0.01, samples_per_gpus=2 and max_epochs=4 will be better.

Too many iterations can lead to overfitting.