Mmdetection: Training doesn't start for multi-gpu on custom dataset

[mdv3101]

Describe the bug
I am training a cascade mask rcnn with dcn on custom dataset (having one class). When i am using single GPU, it is working fine. But when i switch to 2 GPUs, training won't start. Also if I remove custom class, then training start without any problem. Facing such a strange behaviour.

Reproduction

1. What command or script did you run?

./tools/dist_train.sh configs/dcn/cascade_mask_rcnn_r101_fpn_dconv_c3-c5_1x_coco.py 2

1. Did you make any modifications on the code or config? Did you understand what you have modified?
   I added classes in config/_base_/dataset/coco_detection.py
   Also i changed the num_classes=1 in config/_base_/models/cascade_mask_rcnn_r50_fpn.py

Log File Output

2020-06-04 20:04:14,189 - mmdet - INFO - Environment info:
------------------------------------------------------------
sys.platform: linux
Python: 3.7.3 (default, Mar 27 2019, 22:11:17) [GCC 7.3.0]
CUDA available: True
CUDA_HOME: /usr/local/cuda-10.0
NVCC: Cuda compilation tools, release 10.0, V10.0.130
GPU 0,1: GeForce RTX 2080 Ti
GCC: gcc (Ubuntu 5.5.0-12ubuntu1~16.04) 5.5.0 20171010
PyTorch: 1.4.0
PyTorch compiling details: PyTorch built with:
  - GCC 7.3
  - Intel(R) Math Kernel Library Version 2019.0.4 Product Build 20190411 for Intel(R) 64 architecture applications
  - Intel(R) MKL-DNN v0.21.1 (Git Hash 7d2fd500bc78936d1d648ca713b901012f470dbc)
  - OpenMP 201511 (a.k.a. OpenMP 4.5)
  - NNPACK is enabled
  - CUDA Runtime 10.1
  - 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_61,code=sm_61;-gencode;arch=compute_70,code=sm_70;-gencode;arch=compute_75,code=sm_75;-gencode;arch=compute_37,code=compute_37
  - CuDNN 7.6.3
  - Magma 2.5.1
  - Build settings: BLAS=MKL, BUILD_NAMEDTENSOR=OFF, BUILD_TYPE=Release, CXX_FLAGS= -Wno-deprecated -fvisibility-inlines-hidden -fopenmp -DUSE_FBGEMM -DUSE_QNNPACK -DUSE_PYTORCH_QNNPACK -O2 -fPIC -Wno-narrowing -Wall -Wextra -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-strict-overflow -Wno-strict-aliasing -Wno-error=deprecated-declarations -Wno-stringop-overflow -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 -Wno-stringop-overflow, DISABLE_NUMA=1, PERF_WITH_AVX=1, PERF_WITH_AVX2=1, PERF_WITH_AVX512=1, USE_CUDA=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, USE_STATIC_DISPATCH=OFF, 

TorchVision: 0.5.0
OpenCV: 4.2.0
MMCV: 0.5.5
MMDetection: 2.0.0+a9bedfb
MMDetection Compiler: GCC 5.5
MMDetection CUDA Compiler: 10.0
------------------------------------------------------------

2020-06-04 20:04:14,190 - mmdet - INFO - Distributed training: True
2020-06-04 20:04:15,147 - mmdet - INFO - Config:
model = dict(
    type='CascadeRCNN',
    pretrained='torchvision://resnet101',
    backbone=dict(
        type='ResNet',
        depth=101,
        num_stages=4,
        out_indices=(0, 1, 2, 3),
        frozen_stages=1,
        norm_cfg=dict(type='BN', requires_grad=True),
        norm_eval=True,
        style='pytorch',
        dcn=dict(type='DCN', deformable_groups=1, fallback_on_stride=False),
        stage_with_dcn=(False, True, True, True)),
    neck=dict(
        type='FPN',
        in_channels=[256, 512, 1024, 2048],
        out_channels=256,
        num_outs=5),
    rpn_head=dict(
        type='RPNHead',
        in_channels=256,
        feat_channels=256,
        anchor_generator=dict(
            type='AnchorGenerator',
            scales=[8],
            ratios=[0.5, 1.0, 2.0],
            strides=[4, 8, 16, 32, 64]),
        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='SmoothL1Loss', beta=0.1111111111111111, loss_weight=1.0)),
    roi_head=dict(
        type='CascadeRoIHead',
        num_stages=3,
        stage_loss_weights=[1, 0.5, 0.25],
        bbox_roi_extractor=dict(
            type='SingleRoIExtractor',
            roi_layer=dict(type='RoIAlign', out_size=7, sample_num=0),
            out_channels=256,
            featmap_strides=[4, 8, 16, 32]),
        bbox_head=[
            dict(
                type='Shared2FCBBoxHead',
                in_channels=256,
                fc_out_channels=1024,
                roi_feat_size=7,
                num_classes=1,
                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=True,
                loss_cls=dict(
                    type='CrossEntropyLoss',
                    use_sigmoid=False,
                    loss_weight=1.0),
                loss_bbox=dict(type='SmoothL1Loss', beta=1.0,
                               loss_weight=1.0)),
            dict(
                type='Shared2FCBBoxHead',
                in_channels=256,
                fc_out_channels=1024,
                roi_feat_size=7,
                num_classes=1,
                bbox_coder=dict(
                    type='DeltaXYWHBBoxCoder',
                    target_means=[0.0, 0.0, 0.0, 0.0],
                    target_stds=[0.05, 0.05, 0.1, 0.1]),
                reg_class_agnostic=True,
                loss_cls=dict(
                    type='CrossEntropyLoss',
                    use_sigmoid=False,
                    loss_weight=1.0),
                loss_bbox=dict(type='SmoothL1Loss', beta=1.0,
                               loss_weight=1.0)),
            dict(
                type='Shared2FCBBoxHead',
                in_channels=256,
                fc_out_channels=1024,
                roi_feat_size=7,
                num_classes=1,
                bbox_coder=dict(
                    type='DeltaXYWHBBoxCoder',
                    target_means=[0.0, 0.0, 0.0, 0.0],
                    target_stds=[0.033, 0.033, 0.067, 0.067]),
                reg_class_agnostic=True,
                loss_cls=dict(
                    type='CrossEntropyLoss',
                    use_sigmoid=False,
                    loss_weight=1.0),
                loss_bbox=dict(type='SmoothL1Loss', beta=1.0, loss_weight=1.0))
        ],
        mask_roi_extractor=dict(
            type='SingleRoIExtractor',
            roi_layer=dict(type='RoIAlign', out_size=14, sample_num=0),
            out_channels=256,
            featmap_strides=[4, 8, 16, 32]),
        mask_head=dict(
            type='FCNMaskHead',
            num_convs=4,
            in_channels=256,
            conv_out_channels=256,
            num_classes=1,
            loss_mask=dict(
                type='CrossEntropyLoss', use_mask=True, 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_across_levels=False,
        nms_pre=2000,
        nms_post=2000,
        max_num=2000,
        nms_thr=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),
            mask_size=28,
            pos_weight=-1,
            debug=False),
        dict(
            assigner=dict(
                type='MaxIoUAssigner',
                pos_iou_thr=0.6,
                neg_iou_thr=0.6,
                min_pos_iou=0.6,
                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),
            mask_size=28,
            pos_weight=-1,
            debug=False),
        dict(
            assigner=dict(
                type='MaxIoUAssigner',
                pos_iou_thr=0.7,
                neg_iou_thr=0.7,
                min_pos_iou=0.7,
                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),
            mask_size=28,
            pos_weight=-1,
            debug=False)
    ])
test_cfg = dict(
    rpn=dict(
        nms_across_levels=False,
        nms_pre=1000,
        nms_post=1000,
        max_num=1000,
        nms_thr=0.7,
        min_bbox_size=0),
    rcnn=dict(
        score_thr=0.05,
        nms=dict(type='nms', iou_thr=0.5),
        max_per_img=100,
        mask_thr_binary=0.5))
dataset_type = 'CocoDataset'
classes = './class_file.txt'
data_root = '/ssd_scratch/cvit/ajoy/train_dataset/coco/'
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, with_mask=True),
    dict(type='Resize', img_scale=(1333, 800), 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', 'gt_masks'])
]
test_pipeline = [
    dict(type='LoadImageFromFile'),
    dict(
        type='MultiScaleFlipAug',
        img_scale=(1333, 800),
        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=2,
    workers_per_gpu=2,
    train=dict(
        type='CocoDataset',
        classes='./class_file.txt',
        ann_file=
        '/ssd_scratch/cvit/ajoy/train_dataset/coco/annotations/instances_train2014.json',
        img_prefix='/ssd_scratch/cvit/ajoy/train_dataset/coco/train2014/',
        pipeline=[
            dict(type='LoadImageFromFile'),
            dict(type='LoadAnnotations', with_bbox=True, with_mask=True),
            dict(type='Resize', img_scale=(1333, 800), 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', 'gt_masks'])
        ]),
    val=dict(
        type='CocoDataset',
        classes='./class_file.txt',
        ann_file=
        '/ssd_scratch/cvit/ajoy/train_dataset/coco/annotations/instances_val2014.json',
        img_prefix='/ssd_scratch/cvit/ajoy/train_dataset/coco/val2014/',
        pipeline=[
            dict(type='LoadImageFromFile'),
            dict(
                type='MultiScaleFlipAug',
                img_scale=(1333, 800),
                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='CocoDataset',
        classes='./class_file.txt',
        ann_file=
        '/ssd_scratch/cvit/ajoy/train_dataset/coco/annotations/instances_val2014.json',
        img_prefix='/ssd_scratch/cvit/ajoy/train_dataset/coco/val2014/',
        pipeline=[
            dict(type='LoadImageFromFile'),
            dict(
                type='MultiScaleFlipAug',
                img_scale=(1333, 800),
                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=5, metric=['bbox', 'segm'])
optimizer = dict(type='SGD', lr=0.001, momentum=0.9, weight_decay=0.0001)
optimizer_config = dict(grad_clip=None)
lr_config = dict(
    policy='step',
    warmup='linear',
    warmup_iters=500,
    warmup_ratio=0.001,
    step=[25, 40])
total_epochs = 50
checkpoint_config = dict(interval=1)
log_config = dict(interval=50, hooks=[dict(type='TextLoggerHook')])
dist_params = dict(backend='nccl')
log_level = 'INFO'
load_from = '/ssd_scratch/cvit/ajoy/mmdetection/def_model/cascade_mask_rcnn_r101_fpn_dconv_c3-c5_1x_coco_20200204-df0c5f10.pth'
resume_from = None
workflow = [('train', 1)]
work_dir = '/ssd_scratch/cvit/ajoy/train_dataset/coco/logs'
gpu_ids = range(0, 2)

2020-06-04 20:04:19,373 - mmdet - WARNING - The model and loaded state dict do not match exactly

unexpected key in source state_dict: fc.weight, fc.bias

missing keys in source state_dict: layer2.0.conv2.conv_offset.weight, layer2.0.conv2.conv_offset.bias, layer2.1.conv2.conv_offset.weight, layer2.1.conv2.conv_offset.bias, layer2.2.conv2.conv_offset.weight, layer2.2.conv2.conv_offset.bias, layer2.3.conv2.conv_offset.weight, layer2.3.conv2.conv_offset.bias, layer3.0.conv2.conv_offset.weight, layer3.0.conv2.conv_offset.bias, layer3.1.conv2.conv_offset.weight, layer3.1.conv2.conv_offset.bias, layer3.2.conv2.conv_offset.weight, layer3.2.conv2.conv_offset.bias, layer3.3.conv2.conv_offset.weight, layer3.3.conv2.conv_offset.bias, layer3.4.conv2.conv_offset.weight, layer3.4.conv2.conv_offset.bias, layer3.5.conv2.conv_offset.weight, layer3.5.conv2.conv_offset.bias, layer3.6.conv2.conv_offset.weight, layer3.6.conv2.conv_offset.bias, layer3.7.conv2.conv_offset.weight, layer3.7.conv2.conv_offset.bias, layer3.8.conv2.conv_offset.weight, layer3.8.conv2.conv_offset.bias, layer3.9.conv2.conv_offset.weight, layer3.9.conv2.conv_offset.bias, layer3.10.conv2.conv_offset.weight, layer3.10.conv2.conv_offset.bias, layer3.11.conv2.conv_offset.weight, layer3.11.conv2.conv_offset.bias, layer3.12.conv2.conv_offset.weight, layer3.12.conv2.conv_offset.bias, layer3.13.conv2.conv_offset.weight, layer3.13.conv2.conv_offset.bias, layer3.14.conv2.conv_offset.weight, layer3.14.conv2.conv_offset.bias, layer3.15.conv2.conv_offset.weight, layer3.15.conv2.conv_offset.bias, layer3.16.conv2.conv_offset.weight, layer3.16.conv2.conv_offset.bias, layer3.17.conv2.conv_offset.weight, layer3.17.conv2.conv_offset.bias, layer3.18.conv2.conv_offset.weight, layer3.18.conv2.conv_offset.bias, layer3.19.conv2.conv_offset.weight, layer3.19.conv2.conv_offset.bias, layer3.20.conv2.conv_offset.weight, layer3.20.conv2.conv_offset.bias, layer3.21.conv2.conv_offset.weight, layer3.21.conv2.conv_offset.bias, layer3.22.conv2.conv_offset.weight, layer3.22.conv2.conv_offset.bias, layer4.0.conv2.conv_offset.weight, layer4.0.conv2.conv_offset.bias, layer4.1.conv2.conv_offset.weight, layer4.1.conv2.conv_offset.bias, layer4.2.conv2.conv_offset.weight, layer4.2.conv2.conv_offset.bias

2020-06-04 20:04:22,498 - mmdet - INFO - load checkpoint from /ssd_scratch/cvit/ajoy/mmdetection/def_model/cascade_mask_rcnn_r101_fpn_dconv_c3-c5_1x_coco_20200204-df0c5f10.pth
2020-06-04 20:04:22,917 - mmdet - WARNING - The model and loaded state dict do not match exactly

size mismatch for roi_head.bbox_head.0.fc_cls.weight: copying a param with shape torch.Size([81, 1024]) from checkpoint, the shape in current model is torch.Size([2, 1024]).
size mismatch for roi_head.bbox_head.0.fc_cls.bias: copying a param with shape torch.Size([81]) from checkpoint, the shape in current model is torch.Size([2]).
size mismatch for roi_head.bbox_head.1.fc_cls.weight: copying a param with shape torch.Size([81, 1024]) from checkpoint, the shape in current model is torch.Size([2, 1024]).
size mismatch for roi_head.bbox_head.1.fc_cls.bias: copying a param with shape torch.Size([81]) from checkpoint, the shape in current model is torch.Size([2]).
size mismatch for roi_head.bbox_head.2.fc_cls.weight: copying a param with shape torch.Size([81, 1024]) from checkpoint, the shape in current model is torch.Size([2, 1024]).
size mismatch for roi_head.bbox_head.2.fc_cls.bias: copying a param with shape torch.Size([81]) from checkpoint, the shape in current model is torch.Size([2]).
size mismatch for roi_head.mask_head.0.conv_logits.weight: copying a param with shape torch.Size([80, 256, 1, 1]) from checkpoint, the shape in current model is torch.Size([1, 256, 1, 1]).
size mismatch for roi_head.mask_head.0.conv_logits.bias: copying a param with shape torch.Size([80]) from checkpoint, the shape in current model is torch.Size([1]).
size mismatch for roi_head.mask_head.1.conv_logits.weight: copying a param with shape torch.Size([80, 256, 1, 1]) from checkpoint, the shape in current model is torch.Size([1, 256, 1, 1]).
size mismatch for roi_head.mask_head.1.conv_logits.bias: copying a param with shape torch.Size([80]) from checkpoint, the shape in current model is torch.Size([1]).
size mismatch for roi_head.mask_head.2.conv_logits.weight: copying a param with shape torch.Size([80, 256, 1, 1]) from checkpoint, the shape in current model is torch.Size([1, 256, 1, 1]).
size mismatch for roi_head.mask_head.2.conv_logits.bias: copying a param with shape torch.Size([80]) from checkpoint, the shape in current model is torch.Size([1]).
2020-06-04 20:04:22,935 - mmdet - INFO - Start running, host: ajoy.mondal@gnode59, work_dir: /ssd_scratch/cvit/ajoy/train_dataset/coco/logs
2020-06-04 20:04:22,935 - mmdet - INFO - workflow: [('train', 1)], max: 50 epochs

class_file.txt

Strange thing is, same code is working fine on single GPU

[mdv3101]

I was debugging the code, and find out that I have set
find_unused_parameters = cfg.get('find_unused_parameters', True).

It might have to do something with #2153. I have posted my Traceback in that issue as well. There might be some bug in distributed training.
Please anyone help in solving it.

[hellock]

Such things can happen when two GPUs have different paths, e.g., GPU1 only generate RoIs for FPN level 3 and 4, and GPU2 only generates RoIs for FPN level 4 and 5, then it will get stuck during the backpropagation.

[mdv3101]

Hi @hellock,
Thanks for you response. I understood. But why this is happening only when we are using custom classes?
Also is there any solution to solve this problem.

[MichaelXin]

got the very same error.
I train cascade rcnn on a 1-class dataset. It works well on a single GPU but the program crashes somewhile on 4 GPUs.
When I train it on coco dataset using the official configs, I have not observed the same problem.
can not explain it... and need some luck to achieve multi-GPU training
Help!!!