how to test the SP-NAS mAP on COCO with vega

[xieenze]

how to test the SP-NAS mAP on COCO with vega

[chanyn]

how to test the SP-NAS mAP on COCO with vega

how to test the SP-NAS mAP on COCO with vega

Hi, if you want to reproduce the best performance of SP-NAS on COCO, you should get ImageNet pretrained weight of the corresponding model code in the paper (1111-21111-2111111111111111111111111-211-1), and use the following config file to replace config_template in vega/examples/nas/sp_nas/spnas.yml. Uncomment fullytrain, and remove nas1 and nas2. You will get the full training models and the test results.

"""This is the model settings file for SPNet."""
norm_cfg = dict(type='BN', requires_grad=True)
model = dict(
    type='CascadeRCNN',
    num_stages=3,
    pretrained='/efs-hn/pretrained/backbone/SPNetXB_COCO_ImageNetPretrained.pth',
    backbone=dict(
        type='SpNet',
        pretrained_arch='1111-21111-2111111111111111111111111-211-1',
        layers_arch='1111-21111-2111111111111111111111111-211-1',
        mb_arch='1-0-2-1-2',
        reignition=True,
        arch_block='Bottleneck',
        groups=32,    # ResNext block
        base_width=4,    # ResNext block
        num_stages=5,
        strides=(1, 2, 2, 2, 2),
        dilations=(1, 1, 1, 1, 1),
        out_indices=(0, 1, 2, 3, 4),  # if addit_stage=True, num of out_indices + 1
        stage_with_dcn=(False, False, False, False, False),
        stage_with_gcb=(False, False, False, False, False),
        stage_with_gen_attention=((), (), (), (), ()),
        frozen_stages=0,
        norm_cfg=norm_cfg,
        norm_eval=False,
        style='pytorch'),
    neck=dict(
        type='SpFPN',
        in_channels=[256, 512, 1024, 2048, 2048],
        out_channels=256,
        num_outs=6),
    rpn_head=dict(
        type='RPNHead',
        in_channels=256,
        feat_channels=256,
        anchor_scales=[8],
        anchor_ratios=[0.5, 1.0, 2.0],
        anchor_strides=[4, 8, 16, 32, 64, 128],
        target_means=[.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=1.0 / 9.0, loss_weight=1.0)),
    bbox_roi_extractor=dict(
        type='SingleRoIExtractor',
        roi_layer=dict(type='RoIAlign', out_size=7, sample_num=2),
        out_channels=256,
        featmap_strides=[4, 8, 16, 32, 64]),
    bbox_head=[
        dict(
            type='SharedFCBBoxHead',
            num_fcs=2,
            in_channels=256,
            fc_out_channels=1024,
            roi_feat_size=7,
            num_classes=81,
            target_means=[0., 0., 0., 0.],
            target_stds=[0.1, 0.1, 0.2, 0.2],
            reg_class_agnostic=True,
            norm_cfg=norm_cfg,
            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='SharedFCBBoxHead',
            num_fcs=2,
            in_channels=256,
            fc_out_channels=1024,
            roi_feat_size=7,
            num_classes=81,
            target_means=[0., 0., 0., 0.],
            target_stds=[0.05, 0.05, 0.1, 0.1],
            reg_class_agnostic=True,
            norm_cfg=norm_cfg,
            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='SharedFCBBoxHead',
            num_fcs=2,
            in_channels=256,
            fc_out_channels=1024,
            roi_feat_size=7,
            num_classes=81,
            target_means=[0., 0., 0., 0.],
            target_stds=[0.033, 0.033, 0.067, 0.067],
            reg_class_agnostic=True,
            norm_cfg=norm_cfg,
            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=2),
        out_channels=256,
        featmap_strides=[4, 8, 16, 32, 64]),
    mask_head=dict(
        type='FCNMaskHead',
        num_convs=4,
        in_channels=256,
        conv_out_channels=256,
        num_classes=81,
        norm_cfg=norm_cfg,
        loss_mask=dict(
            type='CrossEntropyLoss', use_mask=True, loss_weight=1.0)))
# model training and testing settings
train_cfg = dict(
    rpn=dict(
        assigner=dict(
            type='MaxIoUAssigner',
            pos_iou_thr=0.7,
            neg_iou_thr=0.3,
            min_pos_iou=0.3,
            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,
            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_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)
    # soft-nms is also supported for rcnn testing
    # e.g., nms=dict(type='soft_nms', iou_thr=0.5, min_score=0.05)
)
# dataset settings
dataset_type = 'CocoDataset'
data_root = '/cache/datasets/COCO2017/'
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=(1280, 1280),
         ratio_range=(0.8, 1.2),
         keep_ratio=True),
    dict(type='RandomCrop', crop_size=(1280, 1280)),
    dict(type='RandomFlip', flip_ratio=0.5),
    dict(type='Normalize', **img_norm_cfg),
    dict(type='Pad', size=(1280, 1280)),
    dict(type='DefaultFormatBundle'),
    dict(type='Collect', keys=['img', 'gt_bboxes', 'gt_labels', 'gt_masks']),
]
test_pipeline = [
    dict(type='LoadImageFromFile'),
    dict(
        type='MultiScaleFlipAug',
        img_scale=(1280, 1280),
        flip=False,
        transforms=[
            dict(type='Resize', keep_ratio=True),
            dict(type='RandomFlip'),
            dict(type='Normalize', **img_norm_cfg),
            dict(type='Pad', size=(1280, 1280)),
            dict(type='ImageToTensor', keys=['img']),
            dict(type='Collect', keys=['img']),
        ])
]
data = dict(
    imgs_per_gpu=1,
    workers_per_gpu=4,
    train=dict(
        type=dataset_type,
        ann_file=data_root + 'annotations/instances_train2017.json',
        img_prefix=data_root + 'train2017/',
        pipeline=train_pipeline),
    val=dict(
        type=dataset_type,
        ann_file=data_root + 'annotations/instances_val2017.json',
        img_prefix=data_root + 'val2017/',
        pipeline=test_pipeline),
    test=dict(
        type=dataset_type,
        ann_file=data_root + 'annotations/instances_val2017.json',
        img_prefix=data_root + 'val2017/',
        pipeline=test_pipeline))

# optimizer
optimizer = dict(type='SGD', lr=0.02, momentum=0.9, weight_decay=0.0001)
optimizer_config = dict(grad_clip=dict(max_norm=35, norm_type=2))
# learning policy
lr_config = dict(
    policy='step',
    warmup='linear',
    warmup_iters=15000,
    warmup_ratio=0.002,
    step=[30, 40])
checkpoint_config = dict(interval=1)
# yapf:disable
log_config = dict(
    interval=500,
    hooks=[
        dict(type='TextLoggerHook'),
        # dict(type='TensorboardLoggerHook')
    ])
# yapf:enable
# runtime settings
total_epochs = 50
dist_params = dict(backend='nccl')
log_level = 'INFO'
work_dir = './work_dirs/spnet_coco_cascade_mask'
load_from = None
resume_from = None
workflow = [('train', 1)]

[AyeshaNirma]

@chanyn @zhangjiajin I have two questions

1. Where can I find SPNetXB_COCO_ImageNetPretrained.pth model ?
2. When you say remove "nas1 and nas2", you mean just have simply the filed in the bottom fully_train ? My impression is that the following .yml file should be used to train ? is this correct, please kindly let me know.

general:
    worker:
        timeout: 1000
        gpus_per_job: -1

#pipeline: [nas1, nas2]
pipeline: [nas1, nas2, fullytrain]

nas1:
    pipe_step:
        type: SpNasPipeStep

    search_algorithm:
        type: SpNas
        codec: SpNasCodec
        total_list: 'total_list_s.csv'
        sample_level: 'serial'
        max_sample: 10
        max_optimal: 5
        serial_settings:
            num_mutate: 3
            addstage_ratio: 0.05
            expend_ratio: 0.3
            max_stages: 6
        regnition: False
#        last_search_result:

    search_space:
        type: SearchSpace
        config_template_file: "./nas/sp_nas/cascade_rcnn_r50_fpn_1x.py"
        epoch: 5

    trainer:
        type: SpNasTrainer
        gpus: 8
        debug: False

nas2:
    pipe_step:
        type: SpNasPipeStep

    search_algorithm:
        type: SpNas
        codec: SpNasCodec
        total_list: 'total_list_p.csv'
        sample_level: 'parallel'
        max_sample: 10
        max_optimal: 1
        serial_settings:
        last_search_result: 'total_list_s.csv'
        regnition: False

    search_space:
        type: SearchSpace
        config_template_file: "./nas/sp_nas/cascade_rcnn_r50_fpn_1x.py"
        epoch: 5

    trainer:
        type: SpNasTrainer
        gpus: 8
        debug: False


fullytrain:
     pipe_step:
         type: FullyTrainPipeStep

     trainer:
         type: SpNasTrainer
         gpus: 8
         model_desc_file: 'total_list_p.csv'
# #        model_desc: 'r_111-2111-211111-211_0-0-0-0'
         config_template: "./nas/sp_nas/cascade_rcnn_r50_fpn_1x.py"
         regnition: False
         epoch: 12
         debug: False