Same config but iou of some class get 0 | 什么都没改但是训练过程中有的类别的iou突然变成0了

[Wanttoretire]

I trained swin+upernet_tiny, but this time iou of some classes like wall, fence, pole etc is 0. The loss is reduce normally And I try two gpus and one gpus but it still got 0. five days ago all things works well and today I just change the PORT in dist_train.sh from29500 to 29501

2023-05-27 14:45:52,111 - mmseg - INFO - Iter [32000/160000] lr: 4.800e-05, eta: 17:17:56, time: 0.517, data_time: 0.004, memory: 8331, decode.loss_dice: 0.2293, decode.loss_focal: 0.0107, decode.acc_seg: 89.7323, aux.loss_dice: 0.2794, aux.loss_focal: 0.0131, aux.acc_seg: 87.3850, loss: 0.5325 2023-05-27 14:48:22,418 - mmseg - INFO - per class results: 2023-05-27 14:48:22,420 - mmseg - INFO - +---------------+-------+-------+ | Class | IoU | Acc | +---------------+-------+-------+ | road | 95.93 | 97.79 | | sidewalk | 76.66 | 89.52 | | building | 85.69 | 95.65 | | wall | 0.0 | 0.0 | | fence | 0.0 | 0.0 | | pole | 56.24 | 69.46 | | traffic light | 59.02 | 75.59 | | traffic sign | 71.09 | 78.95 | | vegetation | 89.16 | 95.58 | | terrain | 52.99 | 66.56 | | sky | 92.89 | 97.7 | | person | 73.15 | 84.3 | | rider | 0.0 | 0.0 | | car | 90.28 | 96.09 | | truck | 0.0 | 0.0 | | bus | 0.0 | 0.0 | | train | 0.0 | 0.0 | | motorcycle | 0.0 | 0.0 | | bicycle | 61.69 | 93.37 | +---------------+-------+-------+ 2023-05-27 14:48:22,420 - mmseg - INFO - Summary: 2023-05-27 14:48:22,420 - mmseg - INFO - +-------+-------+-------+ | aAcc | mIoU | mAcc | +-------+-------+-------+ | 92.73 | 47.62 | 54.77 | +-------+-------+-------+

and the config is

sys.platform: linux
Python: 3.8.15 | packaged by conda-forge | (default, Nov 22 2022, 08:49:35) [GCC 10.4.0]
CUDA available: True
GPU 0,1,2,3: GeForce GTX 1080 Ti
CUDA_HOME: /usr/local/cuda
NVCC: Cuda compilation tools, release 10.2, V10.2.8
GCC: gcc (GCC) 7.1.0
PyTorch: 1.8.0
PyTorch compiling details: PyTorch built with:
  - GCC 7.3
  - C++ Version: 201402
  - Intel(R) oneAPI Math Kernel Library Version 2021.4-Product Build 20210904 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 10.2
  - 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 7.6.5
  - Magma 2.5.2
  - Build settings: BLAS_INFO=mkl, BUILD_TYPE=Release, CUDA_VERSION=10.2, CUDNN_VERSION=7.6.5, 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
OpenCV: 4.6.0
MMCV: 1.7.0
MMCV Compiler: GCC 7.3
MMCV CUDA Compiler: 10.2
MMSegmentation: 0.29.1+77dbecd
------------------------------------------------------------

2023-05-27 10:26:12,956 - mmseg - INFO - Distributed training: True
2023-05-27 10:26:13,872 - mmseg - INFO - Config:
norm_cfg = dict(type='SyncBN', requires_grad=True)
backbone_norm_cfg = dict(type='LN', requires_grad=True)
model = dict(
    type='EncoderDecoder',
    pretrained=None,
    backbone=dict(
        type='SwinTransformer',
        pretrain_img_size=224,
        embed_dims=96,
        patch_size=4,
        window_size=7,
        mlp_ratio=4,
        depths=[2, 2, 6, 2],
        num_heads=[3, 6, 12, 24],
        strides=(4, 2, 2, 2),
        out_indices=(0, 1, 2, 3),
        qkv_bias=True,
        qk_scale=None,
        patch_norm=True,
        drop_rate=0.0,
        attn_drop_rate=0.0,
        drop_path_rate=0.3,
        use_abs_pos_embed=False,
        act_cfg=dict(type='GELU'),
        norm_cfg=dict(type='LN', requires_grad=True),
        init_cfg=dict(
            type='Pretrained',
            checkpoint=
            'https://download.openmmlab.com/mmsegmentation/v0.5/pretrain/swin/swin_tiny_patch4_window7_224_20220317-1cdeb081.pth'
        )),
    decode_head=dict(
        type='UPerHead',
        in_channels=[96, 192, 384, 768],
        in_index=[0, 1, 2, 3],
        pool_scales=(1, 2, 3, 6),
        channels=512,
        dropout_ratio=0.1,
        num_classes=19,
        norm_cfg=dict(type='SyncBN', requires_grad=True),
        align_corners=False,
        loss_decode=[
            dict(type='DiceLoss', use_sigmoid=False, loss_weight=1.0),
            dict(type='FocalLoss', loss_weight=1.0)
        ]),
    auxiliary_head=dict(
        type='FCNHead',
        in_channels=384,
        in_index=2,
        channels=256,
        num_convs=1,
        concat_input=False,
        dropout_ratio=0.1,
        num_classes=19,
        norm_cfg=dict(type='SyncBN', requires_grad=True),
        align_corners=False,
        loss_decode=[
            dict(type='DiceLoss', use_sigmoid=False, loss_weight=1.0),
            dict(type='FocalLoss', loss_weight=1.0)
        ]),
    train_cfg=dict(),
    test_cfg=dict(mode='whole'))
dataset_type = 'CityscapesDataset'
data_root = 'data/cityscapes/'
img_norm_cfg = dict(
    mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_rgb=True)
crop_size = (512, 1024)
train_pipeline = [
    dict(type='LoadImageFromFile'),
    dict(type='LoadAnnotations'),
    dict(type='Resize', img_scale=(2048, 1024), ratio_range=(0.5, 2.0)),
    dict(type='RandomCrop', crop_size=(512, 1024), cat_max_ratio=0.75),
    dict(type='RandomFlip', prob=0.5),
    dict(type='PhotoMetricDistortion'),
    dict(
        type='Normalize',
        mean=[123.675, 116.28, 103.53],
        std=[58.395, 57.12, 57.375],
        to_rgb=True),
    dict(type='Pad', size=(512, 1024), pad_val=0, seg_pad_val=255),
    dict(type='DefaultFormatBundle'),
    dict(type='Collect', keys=['img', 'gt_semantic_seg'])
]
test_pipeline = [
    dict(type='LoadImageFromFile'),
    dict(
        type='MultiScaleFlipAug',
        img_scale=(2048, 1024),
        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='ImageToTensor', keys=['img']),
            dict(type='Collect', keys=['img'])
        ])
]
data = dict(
    samples_per_gpu=2,
    workers_per_gpu=1,
    train=dict(
        type='CityscapesDataset',
        data_root='data/cityscapes/',
        img_dir='leftImg8bit/train',
        ann_dir='gtFine/train',
        pipeline=[
            dict(type='LoadImageFromFile'),
            dict(type='LoadAnnotations'),
            dict(
                type='Resize', img_scale=(2048, 1024), ratio_range=(0.5, 2.0)),
            dict(type='RandomCrop', crop_size=(512, 1024), cat_max_ratio=0.75),
            dict(type='RandomFlip', prob=0.5),
            dict(type='PhotoMetricDistortion'),
            dict(
                type='Normalize',
                mean=[123.675, 116.28, 103.53],
                std=[58.395, 57.12, 57.375],
                to_rgb=True),
            dict(type='Pad', size=(512, 1024), pad_val=0, seg_pad_val=255),
            dict(type='DefaultFormatBundle'),
            dict(type='Collect', keys=['img', 'gt_semantic_seg'])
        ]),
    val=dict(
        type='CityscapesDataset',
        data_root='data/cityscapes/',
        img_dir='leftImg8bit/val',
        ann_dir='gtFine/val',
        pipeline=[
            dict(type='LoadImageFromFile'),
            dict(
                type='MultiScaleFlipAug',
                img_scale=(2048, 1024),
                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='ImageToTensor', keys=['img']),
                    dict(type='Collect', keys=['img'])
                ])
        ]),
    test=dict(
        type='CityscapesDataset',
        data_root='data/cityscapes/',
        img_dir='leftImg8bit/val',
        ann_dir='gtFine/val',
        pipeline=[
            dict(type='LoadImageFromFile'),
            dict(
                type='MultiScaleFlipAug',
                img_scale=(2048, 1024),
                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='ImageToTensor', keys=['img']),
                    dict(type='Collect', keys=['img'])
                ])
        ]),
    train_dataloader=dict(samples_per_gpu=1, workers_per_gpu=1, shuffle=True),
    val_dataloader=dict(samples_per_gpu=1, workers_per_gpu=1, shuffle=False),
    test_dataloader=dict(samples_per_gpu=1, workers_per_gpu=1, shuffle=False))
log_config = dict(
    interval=50, hooks=[dict(type='TextLoggerHook', by_epoch=False)])
dist_params = dict(backend='nccl')
log_level = 'INFO'
load_from = None
resume_from = None
workflow = [('train', 1)]
cudnn_benchmark = True
optimizer = dict(
    type='AdamW',
    lr=6e-05,
    betas=(0.9, 0.999),
    weight_decay=0.01,
    paramwise_cfg=dict(
        custom_keys=dict(
            absolute_pos_embed=dict(decay_mult=0.0),
            relative_position_bias_table=dict(decay_mult=0.0),
            norm=dict(decay_mult=0.0))))
optimizer_config = dict()
lr_config = dict(
    policy='poly',
    warmup='linear',
    warmup_iters=1500,
    warmup_ratio=1e-06,
    power=1.0,
    min_lr=0.0,
    by_epoch=False)
runner = dict(type='IterBasedRunner', max_iters=160000)
checkpoint_config = dict(by_epoch=False, interval=16000)
evaluation = dict(interval=16000, metric='mIoU', pre_eval=True)
checkpoint_file = 'https://download.openmmlab.com/mmsegmentation/v0.5/pretrain/swin/swin_tiny_patch4_window7_224_20220317-1cdeb081.pth'
work_dir = 'my_citscap_tiny/0526tiny_focal_dice'
gpu_ids = range(0, 2)
auto_resume = False

[Falcons95]

I encontour the same problem, same classes got 0 IOU. I ran the experiment using default fcn-unet config, even fine-tuning from model provided in model zoo after a few itertaions will bring IOU of certain class close to 0.

[Wanttoretire]

I encontour the same problem, same classes got 0 IOU. I ran the experiment using default fcn-unet config, even fine-tuning from model provided in model zoo after a few itertaions will bring IOU of certain class close to 0.

That's reaaaally wired. I even ran it on my own computer instead of the server. If u solve the problem plz tell me how TAT

[huanruizhang123]

我好像也遇到了这个问题，很奇怪

[JQ-zhang-sketch]

I also encontour the same problem,but don't know where is wrong.

[joljal]

你有解决这个问题吗？我完全不知道该怎么办。

[Wanttoretire]

I realize what's wrong with my code. It's because I change my loss function from crossentropy to dice loss, but I don't know why iou of some classes become 0 when use dice loss. I also tried git clone the whole git and train many times. It is wired that iou show agin. 我英语太烂了简而言之就是我把损失函数换成了dice loss然后好多iou就没了，换回去也没有用，但是我重新git clone了一下整个代码，然后每500次就eval一次，不知道为啥试了好多次之后iou自己出现了