search

open-mmlab / mmdetection

OpenMMLab Detection Toolbox and Benchmark
https://mmdetection.readthedocs.io
Apache License 2.0
29.36k stars 9.42k forks source link

Bad scores in YOLOX but decent in other architectures #7368

Closed sarmientoj24 closed 2 years ago

sarmientoj24 commented 2 years ago

I am trying to use YoloX as one of the architectures for Object Detection. But compared to other architectures, I am having really bad scores. Any suggestions why?

yolox (last 2 epochs)

[>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>] 351/351, 57.5 task/s, elapsed: 6s, ETA:     0s
---------------iou_thr: 0.5---------------
2022-03-10 10:12:17,974 - mmdet - INFO - 
+--------+-----+------+--------+-------+
| class  | gts | dets | recall | ap    |
+--------+-----+------+--------+-------+
| cavity | 490 | 1792 | 0.080  | 0.002 |
| pa     | 172 | 535  | 0.151  | 0.007 |
+--------+-----+------+--------+-------+
| mAP    |     |      |        | 0.005 |
+--------+-----+------+--------+-------+
2022-03-10 10:12:17,975 - mmdet - INFO - Epoch(val) [97][351]   AP50: 0.0050, mAP: 0.0047
2022-03-10 10:12:28,027 - mmdet - INFO - Epoch [98][50/208]     lr: 6.250e-05, eta: 0:01:40, time: 0.201, data_time: 0.057, memory: 6214, loss_cls: 0.6502, loss_bbox: 2.8659, loss_obj: 2.9006, loss_l1: 0.9162, loss: 7.3329
2022-03-10 10:12:35,786 - mmdet - INFO - Epoch [98][100/208]    lr: 6.250e-05, eta: 0:01:31, time: 0.155, data_time: 0.010, memory: 6214, loss_cls: 0.6269, loss_bbox: 2.7124, loss_obj: 1.7779, loss_l1: 0.7883, loss: 5.9056
2022-03-10 10:12:43,249 - mmdet - INFO - Epoch [98][150/208]    lr: 6.250e-05, eta: 0:01:22, time: 0.149, data_time: 0.009, memory: 6214, loss_cls: 0.6364, loss_bbox: 2.8063, loss_obj: 2.0539, loss_l1: 0.8438, loss: 6.3405
2022-03-10 10:12:50,357 - mmdet - INFO - Epoch [98][200/208]    lr: 6.250e-05, eta: 0:01:13, time: 0.142, data_time: 0.009, memory: 6214, loss_cls: 0.6495, loss_bbox: 2.8711, loss_obj: 2.4804, loss_l1: 0.8849, loss: 6.8859
[>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>] 351/351, 57.5 task/s, elapsed: 6s, ETA:     0s
---------------iou_thr: 0.5---------------
2022-03-10 10:12:58,097 - mmdet - INFO - 
+--------+-----+------+--------+-------+
| class  | gts | dets | recall | ap    |
+--------+-----+------+--------+-------+
| cavity | 490 | 1771 | 0.080  | 0.002 |
| pa     | 172 | 525  | 0.151  | 0.008 |
+--------+-----+------+--------+-------+
| mAP    |     |      |        | 0.005 |
+--------+-----+------+--------+-------+
2022-03-10 10:12:58,099 - mmdet - INFO - Epoch(val) [98][351]   AP50: 0.0050, mAP: 0.0048
2022-03-10 10:13:08,456 - mmdet - INFO - Epoch [99][50/208]     lr: 6.250e-05, eta: 0:01:03, time: 0.207, data_time: 0.056, memory: 6214, loss_cls: 0.6240, loss_bbox: 2.6947, loss_obj: 1.9800, loss_l1: 0.8009, loss: 6.0996
2022-03-10 10:13:16,377 - mmdet - INFO - Epoch [99][100/208]    lr: 6.250e-05, eta: 0:00:55, time: 0.158, data_time: 0.009, memory: 6214, loss_cls: 0.6312, loss_bbox: 2.7407, loss_obj: 1.6332, loss_l1: 0.8054, loss: 5.8105
2022-03-10 10:13:24,525 - mmdet - INFO - Epoch [99][150/208]    lr: 6.250e-05, eta: 0:00:46, time: 0.163, data_time: 0.009, memory: 6214, loss_cls: 0.6387, loss_bbox: 2.7241, loss_obj: 2.0790, loss_l1: 0.8505, loss: 6.2923
2022-03-10 10:13:32,176 - mmdet - INFO - Epoch [99][200/208]    lr: 6.250e-05, eta: 0:00:37, time: 0.153, data_time: 0.009, memory: 6214, loss_cls: 0.6375, loss_bbox: 2.7981, loss_obj: 2.1879, loss_l1: 0.8473, loss: 6.4708
[>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>] 351/351, 57.2 task/s, elapsed: 6s, ETA:     0s
---------------iou_thr: 0.5---------------
2022-03-10 10:13:39,909 - mmdet - INFO - 
+--------+-----+------+--------+-------+
| class  | gts | dets | recall | ap    |
+--------+-----+------+--------+-------+
| cavity | 490 | 1751 | 0.084  | 0.002 |
| pa     | 172 | 514  | 0.151  | 0.008 |
+--------+-----+------+--------+-------+
| mAP    |     |      |        | 0.005 |
+--------+-----+------+--------+-------+
2022-03-10 10:13:39,910 - mmdet - INFO - Epoch(val) [99][351]   AP50: 0.0050, mAP: 0.0050
2022-03-10 10:13:50,469 - mmdet - INFO - Epoch [100][50/208]    lr: 6.250e-05, eta: 0:00:27, time: 0.211, data_time: 0.057, memory: 6214, loss_cls: 0.6431, loss_bbox: 2.8128, loss_obj: 2.2370, loss_l1: 0.8816, loss: 6.5745
2022-03-10 10:13:58,238 - mmdet - INFO - Epoch [100][100/208]   lr: 6.250e-05, eta: 0:00:18, time: 0.155, data_time: 0.009, memory: 6214, loss_cls: 0.6229, loss_bbox: 2.6683, loss_obj: 1.8179, loss_l1: 0.7822, loss: 5.8914
2022-03-10 10:14:05,879 - mmdet - INFO - Epoch [100][150/208]   lr: 6.250e-05, eta: 0:00:10, time: 0.153, data_time: 0.009, memory: 6214, loss_cls: 0.6315, loss_bbox: 2.7117, loss_obj: 1.8303, loss_l1: 0.7885, loss: 5.9621
2022-03-10 10:14:12,759 - mmdet - INFO - Epoch [100][200/208]   lr: 6.250e-05, eta: 0:00:01, time: 0.138, data_time: 0.009, memory: 6214, loss_cls: 0.6424, loss_bbox: 2.8237, loss_obj: 2.3767, loss_l1: 0.8779, loss: 6.7207

retinanet

[>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>] 351/351, 40.0 task/s, elapsed: 9s, ETA:     0s
---------------iou_thr: 0.5---------------
2022-03-10 08:00:48,057 - mmdet - INFO - 
+--------+-----+------+--------+-------+
| class  | gts | dets | recall | ap    |
+--------+-----+------+--------+-------+
| cavity | 490 | 3368 | 0.633  | 0.361 |
| pa     | 172 | 514  | 0.657  | 0.463 |
+--------+-----+------+--------+-------+
| mAP    |     |      |        | 0.412 |
+--------+-----+------+--------+-------+
2022-03-10 08:00:48,058 - mmdet - INFO - Epoch(val) [99][351]   AP50: 0.4120, mAP: 0.4119
2022-03-10 08:01:12,034 - mmdet - INFO - Epoch [100][50/104]    lr: 1.250e-05, eta: 0:00:23, time: 0.479, data_time: 0.095, memory: 5343, loss_cls: 0.1659, loss_bbox: 0.2610, loss: 0.4270, grad_norm: 5.2169
2022-03-10 08:01:34,911 - mmdet - INFO - Epoch [100][100/104]   lr: 1.250e-05, eta: 0:00:01, time: 0.458, data_time: 0.035, memory: 5343, loss_cls: 0.1818, loss_bbox: 0.2742, loss: 0.4560, grad_norm: 5.6148
2022-03-10 08:01:36,928 - mmdet - INFO - Saving checkpoint at 100 epochs
[>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>] 351/351, 25.2 task/s, elapsed: 14s, ETA:     0s
---------------iou_thr: 0.5---------------
2022-03-10 08:01:52,669 - mmdet - INFO - 
+--------+-----+------+--------+-------+
| class  | gts | dets | recall | ap    |
+--------+-----+------+--------+-------+
| cavity | 490 | 3298 | 0.635  | 0.363 |
| pa     | 172 | 495  | 0.680  | 0.496 |
+--------+-----+------+--------+-------+
| mAP    |     |      |        | 0.429 |
+--------+-----+------+--------+-------+
2022-03-10 08:01:52,672 - mmdet - INFO - Epoch(val) [100][351]  AP50: 0.4290, mAP: 0.4294

Cascade RCNN

[>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>] 351/351, 20.2 task/s, elapsed: 17s, ETA:     0s
---------------iou_thr: 0.5---------------
2022-03-10 08:20:16,910 - mmdet - INFO - 
+--------+-----+------+--------+-------+
| class  | gts | dets | recall | ap    |
+--------+-----+------+--------+-------+
| cavity | 490 | 898  | 0.504  | 0.326 |
| pa     | 172 | 264  | 0.610  | 0.464 |
+--------+-----+------+--------+-------+
| mAP    |     |      |        | 0.395 |
+--------+-----+------+--------+-------+
2022-03-10 08:20:16,913 - mmdet - INFO - Epoch(val) [99][351]   AP50: 0.3950, mAP: 0.3950
2022-03-10 08:21:14,781 - mmdet - INFO - Epoch [100][50/104]    lr: 2.500e-05, eta: 0:00:56, time: 1.155, data_time: 0.108, memory: 7542, loss_rpn_cls: 0.0114, loss_rpn_bbox: 0.0054, s0.loss_cls: 0.0579, s0.acc: 97.7305, s0.loss_bbox: 0.0477, s1.loss_cls: 0.0298, s1.acc: 97.6778, s1.loss_bbox: 0.0626, s2.loss_cls: 0.0156, s2.acc: 97.4595, s2.loss_bbox: 0.0350, loss: 0.2655
2022-03-10 08:22:07,842 - mmdet - INFO - Epoch [100][100/104]   lr: 2.500e-05, eta: 0:00:04, time: 1.061, data_time: 0.033, memory: 7542, loss_rpn_cls: 0.0145, loss_rpn_bbox: 0.0062, s0.loss_cls: 0.0560, s0.acc: 97.8596, s0.loss_bbox: 0.0450, s1.loss_cls: 0.0293, s1.acc: 97.7130, s1.loss_bbox: 0.0600, s2.loss_cls: 0.0152, s2.acc: 97.4996, s2.loss_bbox: 0.0330, loss: 0.2593
2022-03-10 08:22:12,175 - mmdet - INFO - Saving checkpoint at 100 epochs
[>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>] 351/351, 24.6 task/s, elapsed: 14s, ETA:     0s
---------------iou_thr: 0.5---------------
2022-03-10 08:22:28,972 - mmdet - INFO - 
+--------+-----+------+--------+-------+
| class  | gts | dets | recall | ap    |
+--------+-----+------+--------+-------+
| cavity | 490 | 879  | 0.496  | 0.322 |
| pa     | 172 | 262  | 0.616  | 0.466 |
+--------+-----+------+--------+-------+
| mAP    |     |      |        | 0.394 |
+--------+-----+------+--------+-------+
2022-03-10 08:22:28,974 - mmdet - INFO - Epoch(val) [100][351]  AP50: 0.3940, mAP: 0.3940

Faster RCNN

2022-03-10 07:06:27,541 - mmdet - INFO -
+--------+-----+------+--------+-------+
| class  | gts | dets | recall | ap    |
+--------+-----+------+--------+-------+
| cavity | 490 | 989  | 0.551  | 0.349 |
| pa     | 172 | 306  | 0.587  | 0.439 |
+--------+-----+------+--------+-------+
| mAP    |     |      |        | 0.394 |
+--------+-----+------+--------+-------+
2022-03-10 07:06:27,543 - mmdet - INFO - Epoch(val) [99][351] AP50: 0.3940, mAP: 0.3941
2022-03-10 07:07:07,670 - mmdet - INFO - Epoch [100][50/104] lr: 2.500e-05, eta: 0:00:36, time: 0.800, data_time: 0.104, memory: 6896, loss_rpn_cls: 0.0099, loss_rpn_bbox: 0.0093, loss_cls: 0.0730, acc: 97.1970, loss_bbox: 0.1103, loss: 0.2025
2022-03-10 07:07:44,571 - mmdet - INFO - Epoch [100][100/104] lr: 2.500e-05, eta: 0:00:02, time: 0.738, data_time: 0.031, memory: 6896, loss_rpn_cls: 0.0125, loss_rpn_bbox: 0.0102, loss_cls: 0.0710, acc: 97.3193, loss_bbox: 0.1049, loss: 0.1986
2022-03-10 07:07:47,327 - mmdet - INFO - Saving checkpoint at 100 epochs
[>>>>>>>>>>                                        ] 74/351, 25.9 task/s, elapsed: 3s, ETA:    11s0s
[>>>>>>>>>>>>>>>>>>>>                              ] 145/351, 29.6 task/s, elapsed: 5s, ETA:     7ss
[>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>                    ] 213/351, 30.8 task/s, elapsed: 7s, ETA:     4ss
[>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>          ] 282/351, 31.4 task/s, elapsed: 9s, ETA:     2ss
[>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>> ] 345/351, 31.4 task/s, elapsed: 11s, ETA:     0s
---------------iou_thr: 0.5--------------->>>>>>>>>] 351/351, 31.4 task/s, elapsed: 11s, ETA:     0s
---------------iou_thr: 0.5--------------->>>>>>>>>] 351/351, 31.4 task/s, elapsed: 11s, ETA:     0s
---------------iou_thr: 0.5--------------->>>>>>>>>] 351/351, 31.4 task/s, elapsed: 11s, ETA:     0s
2022-03-10 07:08:01,956 - mmdet - INFO -
+--------+-----+------+--------+-------+
| class  | gts | dets | recall | ap    |
+--------+-----+------+--------+-------+
| cavity | 490 | 966  | 0.555  | 0.352 |
| pa     | 172 | 304  | 0.587  | 0.438 |
+--------+-----+------+--------+-------+
| mAP    |     |      |        | 0.395 |
+--------+-----+------+--------+-------+
2022-03-10 07:08:01,957 - mmdet - INFO - Epoch(val) [100][351] AP50: 0.3950, mAP: 0.3947

YOLOX Config

Config:
optimizer = dict(
    type='SGD',
    lr=0.00125,
    momentum=0.9,
    weight_decay=0.0005,
    nesterov=True,
    paramwise_cfg=dict(norm_decay_mult=0.0, bias_decay_mult=0.0))
optimizer_config = dict(grad_clip=None)
lr_config = dict(
    policy='YOLOX',
    warmup='exp',
    by_epoch=False,
    warmup_by_epoch=True,
    warmup_ratio=1,
    warmup_iters=5,
    num_last_epochs=15,
    min_lr_ratio=0.05)
runner = dict(type='EpochBasedRunner', max_epochs=100)
checkpoint_config = dict(interval=20)
log_config = dict(
    interval=50,
    hooks=[
        dict(type='TextLoggerHook'),
        dict(
            type='WandbLoggerHook',
            init_kwargs=dict(
                project='adra_bipa_mmdet',
                name='yolox-512-conf-001-noaugs',
                config=dict(work_dirs='./logs_bipa_yoloxv1')),
            by_epoch=True)
    ])
custom_hooks = [
    dict(type='YOLOXModeSwitchHook', num_last_epochs=15, priority=48),
    dict(type='SyncNormHook', num_last_epochs=15, interval=10, priority=48),
    dict(
        type='ExpMomentumEMAHook',
        resume_from=None,
        momentum=0.0001,
        priority=49)
]
dist_params = dict(backend='nccl')
log_level = 'INFO'
load_from = 'checkpoints/yolox_s_8x8_300e_coco_20211121_095711-4592a793.pth'
resume_from = None
workflow = [('train', 1)]
opencv_num_threads = 0
mp_start_method = 'fork'
img_scale = (512, 512)
model = dict(
    type='YOLOX',
    input_size=(640, 640),
    random_size_range=(15, 25),
    random_size_interval=10,
    backbone=dict(type='CSPDarknet', deepen_factor=0.33, widen_factor=0.5),
    neck=dict(
        type='YOLOXPAFPN',
        in_channels=[128, 256, 512],
        out_channels=128,
        num_csp_blocks=1),
    bbox_head=dict(
        type='YOLOXHead', num_classes=2, in_channels=128, feat_channels=128),
    train_cfg=dict(assigner=dict(type='SimOTAAssigner', center_radius=2.5)),
    test_cfg=dict(score_thr=0.01, nms=dict(type='nms', iou_threshold=0.4)))
data_root = 'dataset/bipa'
dataset_type = 'CocoDataset'
train_pipeline = [
    dict(type='Mosaic', img_scale=(512, 512), pad_val=114.0),
    dict(
        type='RandomAffine', scaling_ratio_range=(0.1, 2),
        border=(-256, -256)),
    dict(
        type='MixUp',
        img_scale=(512, 512),
        ratio_range=(0.8, 1.6),
        pad_val=114.0),
    dict(type='YOLOXHSVRandomAug'),
    dict(type='RandomFlip', flip_ratio=0.0),
    dict(type='Resize', img_scale=(512, 512), keep_ratio=True),
    dict(
        type='Pad',
        pad_to_square=True,
        pad_val=dict(img=(114.0, 114.0, 114.0))),
    dict(type='FilterAnnotations', min_gt_bbox_wh=(1, 1), keep_empty=False),
    dict(type='DefaultFormatBundle'),
    dict(type='Collect', keys=['img', 'gt_bboxes', 'gt_labels'])
]
train_dataset = dict(
    type='MultiImageMixDataset',
    dataset=dict(
        type='BIPADataset',
        ann_file='dataset/bipa/annotations/train.json',
        img_prefix='dataset/bipa/train',
        pipeline=[
            dict(type='LoadImageFromFile'),
            dict(type='LoadAnnotations', with_bbox=True)
        ],
        filter_empty_gt=False),
    pipeline=[
        dict(type='Mosaic', img_scale=(512, 512), pad_val=114.0),
        dict(
            type='RandomAffine',
            scaling_ratio_range=(0.1, 2),
            border=(-256, -256)),
        dict(
            type='MixUp',
            img_scale=(512, 512),
            ratio_range=(0.8, 1.6),
            pad_val=114.0),
        dict(type='YOLOXHSVRandomAug'),
        dict(type='RandomFlip', flip_ratio=0.0),
        dict(type='Resize', img_scale=(512, 512), keep_ratio=True),
        dict(
            type='Pad',
            pad_to_square=True,
            pad_val=dict(img=(114.0, 114.0, 114.0))),
        dict(
            type='FilterAnnotations', min_gt_bbox_wh=(1, 1), keep_empty=False),
        dict(type='DefaultFormatBundle'),
        dict(type='Collect', keys=['img', 'gt_bboxes', 'gt_labels'])
    ])
test_pipeline = [
    dict(type='LoadImageFromFile'),
    dict(
        type='MultiScaleFlipAug',
        img_scale=(512, 512),
        flip=False,
        transforms=[
            dict(type='Resize', keep_ratio=True),
            dict(type='RandomFlip'),
            dict(
                type='Pad',
                pad_to_square=True,
                pad_val=dict(img=(114.0, 114.0, 114.0))),
            dict(type='DefaultFormatBundle'),
            dict(type='Collect', keys=['img'])
        ])
]
data = dict(
    samples_per_gpu=8,
    workers_per_gpu=8,
    persistent_workers=True,
    train=dict(
        type='MultiImageMixDataset',
        dataset=dict(
            type='BIPADataset',
            ann_file='dataset/bipa/annotations/train.json',
            img_prefix='dataset/bipa/train',
            pipeline=[
                dict(type='LoadImageFromFile'),
                dict(type='LoadAnnotations', with_bbox=True)
            ],
            filter_empty_gt=False),
        pipeline=[
            dict(type='Mosaic', img_scale=(512, 512), pad_val=114.0),
            dict(
                type='RandomAffine',
                scaling_ratio_range=(0.1, 2),
                border=(-256, -256)),
            dict(
                type='MixUp',
                img_scale=(512, 512),
                ratio_range=(0.8, 1.6),
                pad_val=114.0),
            dict(type='YOLOXHSVRandomAug'),
            dict(type='RandomFlip', flip_ratio=0.0),
            dict(type='Resize', img_scale=(512, 512), keep_ratio=True),
            dict(
                type='Pad',
                pad_to_square=True,
                pad_val=dict(img=(114.0, 114.0, 114.0))),
            dict(
                type='FilterAnnotations',
                min_gt_bbox_wh=(1, 1),
                keep_empty=False),
            dict(type='DefaultFormatBundle'),
            dict(type='Collect', keys=['img', 'gt_bboxes', 'gt_labels'])
        ]),
    val=dict(
        type='BIPADataset',
        ann_file='dataset/bipa/annotations/val.json',
        img_prefix='dataset/bipa/val',
        pipeline=[
            dict(type='LoadImageFromFile'),
            dict(
                type='MultiScaleFlipAug',
                img_scale=(512, 512),
                flip=False,
                transforms=[
                    dict(type='Resize', keep_ratio=True),
                    dict(type='RandomFlip'),
                    dict(
                        type='Pad',
                        pad_to_square=True,
                        pad_val=dict(img=(114.0, 114.0, 114.0))),
                    dict(type='DefaultFormatBundle'),
                    dict(type='Collect', keys=['img'])
                ])
        ]),
    test=dict(
        type='BIPADataset',
        ann_file='dataset/bipa/annotations/val.json',
        img_prefix='dataset/bipa/val',
        pipeline=[
            dict(type='LoadImageFromFile'),
            dict(
                type='MultiScaleFlipAug',
                img_scale=(512, 512),
                flip=False,
                transforms=[
                    dict(type='Resize', keep_ratio=True),
                    dict(type='RandomFlip'),
                    dict(
                        type='Pad',
                        pad_to_square=True,
                        pad_val=dict(img=(114.0, 114.0, 114.0))),
                    dict(type='DefaultFormatBundle'),
                    dict(type='Collect', keys=['img'])
                ])
        ]))
max_epochs = 100
num_last_epochs = 5
interval = 10
evaluation = dict(interval=1, dynamic_intervals=[(285, 1)], metric='mAP')
albu_train_transforms = [
    dict(
        type='ShiftScaleRotate',
        shift_limit=0.0625,
        scale_limit=0.2,
        rotate_limit=45,
        interpolation=1,
        p=0.1),
    dict(
        type='RandomBrightnessContrast',
        brightness_limit=[-0.15, 0.15],
        contrast_limit=[-0.15, 0.15],
        p=0.7),
    dict(type='ImageCompression', quality_lower=85, quality_upper=95, p=0.2),
    dict(
        type='OneOf',
        transforms=[
            dict(type='Blur', blur_limit=3, p=1.0),
            dict(type='MedianBlur', blur_limit=3, p=1.0),
            dict(type='GaussianBlur', blur_limit=3, p=1.0)
        ],
        p=0.25),
    dict(
        type='OneOf',
        transforms=[dict(type='Sharpen', p=1.0),
                    dict(type='Emboss', p=1.0)],
        p=0.25),
    dict(type='GaussNoise', var_limit=[20.0, 80.0], per_channel=False, p=0.3),
    dict(type='HorizontalFlip', p=0.5),
    dict(type='RandomRotate90', p=0.3),
    dict(type='Transpose', p=0.2)
]
classes = ('cavity', 'pa')
CLASSES = ('cavity', 'pa')
work_dir = './logs_bipa_yoloxv1'
seed = 0
gpu_ids = [3]
RangiLyu commented 2 years ago

YOLOX uses extremely heavy data augmentation which is adjusted on the coco dataset. And its training pipeline may not be suitable for other datasets.