search

AILab-CVC / YOLO-World

[CVPR 2024] Real-Time Open-Vocabulary Object Detection
https://www.yoloworld.cc
GNU General Public License v3.0
4.64k stars 449 forks source link

The testing results of the whole dataset is empty. (微调不成功） #258

Open mandyxiaomeng opened 6 months ago

mandyxiaomeng commented 6 months ago

您好，

我用了以下config在我自己的数据上微调：

_base_ = (
    '../../third_party/mmyolo/configs/yolov8/'
    'yolov8_l_mask-refine_syncbn_fast_8xb16-500e_coco.py')
custom_imports = dict(
    imports=['yolo_world'],
    allow_failed_imports=False)

# hyper-parameters
class_name = ("couplinghalf","cross", "gear1", "gear2", "pin1", "pin2", "pinion" )
num_classes = len(class_name)
metainfo = dict(classes=class_name, palette=[(220, 20, 60), (150,20,30), (20,220,50), (10,100,100), (200,0,50), (100,50,60), (40,200,40)])
num_training_classes = len(class_name)
max_epochs = 5  # Maximum training epochs
close_mosaic_epochs = 1
save_epoch_intervals = 5

text_channels = 512
neck_embed_channels = [128, 256, _base_.last_stage_out_channels // 2]
neck_num_heads = [4, 8, _base_.last_stage_out_channels // 2 // 32]
base_lr = 2e-4
weight_decay = 0.05
train_batch_size_per_gpu = 16
load_from='pretrained_models/yolo_world_l_clip_base_dual_vlpan_2e-3adamw_32xb16_100e_o365_goldg_train_pretrained-0e566235.pth'
persistent_workers = False

# model settings
model = dict(
    type='YOLOWorldDetector',
    mm_neck=True,
    num_train_classes=num_training_classes,
    num_test_classes=num_classes,
    data_preprocessor=dict(type='YOLOWDetDataPreprocessor'),
    backbone=dict(
        _delete_=True,
        type='MultiModalYOLOBackbone',
        image_model={{_base_.model.backbone}},
        text_model=dict(
            type='HuggingCLIPLanguageBackbone',
            model_name='openai/clip-vit-base-patch32',
            frozen_modules=['all'])),
    neck=dict(type='YOLOWorldDualPAFPN',
              guide_channels=text_channels,
              embed_channels=neck_embed_channels,
              num_heads=neck_num_heads,
              block_cfg=dict(type='MaxSigmoidCSPLayerWithTwoConv'),
              text_enhancder=dict(type='ImagePoolingAttentionModule',
                                  embed_channels=256,
                                  num_heads=8)),
    bbox_head=dict(type='YOLOWorldHead',
                   head_module=dict(type='YOLOWorldHeadModule',
                                    embed_dims=text_channels,
                                    num_classes=num_training_classes)),
    train_cfg=dict(assigner=dict(num_classes=num_training_classes)))

# dataset settings
text_transform = [
    dict(type='RandomLoadText',
         num_neg_samples=(num_classes, num_classes),
         max_num_samples=num_training_classes,
         padding_to_max=True,
         padding_value=''),
    dict(type='mmdet.PackDetInputs',
         meta_keys=('img_id', 'img_path', 'ori_shape', 'img_shape', 'flip',
                    'flip_direction', 'texts'))
]
mosaic_affine_transform = [
    dict(
        type='MultiModalMosaic',
        img_scale=_base_.img_scale,
        pad_val=114.0,
        pre_transform=_base_.pre_transform),
    dict(type='YOLOv5CopyPaste', prob=_base_.copypaste_prob),
    dict(
        type='YOLOv5RandomAffine',
        max_rotate_degree=0.0,
        max_shear_degree=0.0,
        max_aspect_ratio=100.,
        scaling_ratio_range=(1 - _base_.affine_scale,
                             1 + _base_.affine_scale),
        # img_scale is (width, height)
        border=(-_base_.img_scale[0] // 2, -_base_.img_scale[1] // 2),
        border_val=(114, 114, 114),
        min_area_ratio=_base_.min_area_ratio,
        use_mask_refine=_base_.use_mask2refine)
]
train_pipeline = [
    *_base_.pre_transform,
    *mosaic_affine_transform,
    dict(
        type='YOLOv5MultiModalMixUp',
        prob=_base_.mixup_prob,
        pre_transform=[*_base_.pre_transform,
                       *mosaic_affine_transform]),
    *_base_.last_transform[:-1],
    *text_transform
]
train_pipeline_stage2 = [
    *_base_.train_pipeline_stage2[:-1],
    *text_transform
]

coco_train_dataset = dict(
    _delete_=True,
    type='MultiModalDataset',
    dataset=dict(
        type='YOLOv5CocoDataset',
        metainfo=metainfo,
        data_root='data_usecase3_yoloworld',
        ann_file='annotations/train.json',
        data_prefix=dict(img='train/'),
        filter_cfg=dict(filter_empty_gt=False, min_size=32)),
    class_text_path='data/texts/use_case3.json',
    pipeline=train_pipeline)

train_dataloader = dict(
    persistent_workers=persistent_workers,
    batch_size=train_batch_size_per_gpu,
    collate_fn=dict(type='yolow_collate'),
    dataset=coco_train_dataset)

test_pipeline = [
    *_base_.test_pipeline[:-1],
    dict(type='LoadText'),
    dict(
        type='mmdet.PackDetInputs',
        meta_keys=('img_id', 'img_path', 'ori_shape', 'img_shape',
                   'scale_factor', 'pad_param', 'texts'))
]

coco_val_dataset = dict(
    _delete_=True,
    type='MultiModalDataset',
    dataset=dict(
        type='YOLOv5CocoDataset',
        metainfo=metainfo,
        data_root='data_usecase3_yoloworld',
        test_mode=True,
        ann_file='annotations/val.json',
        data_prefix=dict(img='val'),
        filter_cfg=dict(filter_empty_gt=False, min_size=32)),
    class_text_path='data/texts/use_case3.json',
    pipeline=test_pipeline)

val_dataloader = dict(dataset=coco_val_dataset)

test_dataloader = val_dataloader

# training settings
default_hooks = dict(
    param_scheduler=dict(
        scheduler_type='linear',
        lr_factor=0.01,
        max_epochs=max_epochs),
    checkpoint=dict(
        max_keep_ckpts=-1,
        save_best=None,
        interval=save_epoch_intervals))
custom_hooks = [
    dict(
        type='EMAHook',
        ema_type='ExpMomentumEMA',
        momentum=0.0001,
        update_buffers=True,
        strict_load=False,
        priority=49),
    dict(
        type='mmdet.PipelineSwitchHook',
        switch_epoch=max_epochs - close_mosaic_epochs,
        switch_pipeline=train_pipeline_stage2)
]
train_cfg = dict(
    max_epochs=max_epochs,
    val_interval=5,
    dynamic_intervals=[((max_epochs - close_mosaic_epochs),
                        _base_.val_interval_stage2)])
optim_wrapper = dict(
    optimizer=dict(
        _delete_=True,
        type='AdamW',
        lr=base_lr,
        weight_decay=weight_decay,
        batch_size_per_gpu=train_batch_size_per_gpu),
    paramwise_cfg=dict(
        bias_decay_mult=0.0,
        norm_decay_mult=0.0,
        custom_keys={'backbone.text_model': dict(lr_mult=0.01),
                     'logit_scale': dict(weight_decay=0.0)}),
    constructor='YOLOWv5OptimizerConstructor')

# evaluation settings
val_evaluator = dict(
    _delete_=True,
    type='mmdet.CocoMetric',
    proposal_nums=(100, 1, 10),
    ann_file='data_usecase3_yoloworld/annotations/val.json',
    metric='bbox')

#, sort_categories=True, format_only=False

test_evaluator = val_evaluator

我可以实现微调，试着训练了5个epoch得到了epoch_5.pth 然后我想用test.sh看这个模型的测试结果，用了以下命令：

./tools/dist_test.sh ./configs/finetune_coco/yolo_world_l_usecase3.py ./work_dirs/yolo_world_l_usecase3/epoch_5.pth 2 --show

可是show 显示出来的图片没有bounding box, 连GroundTruth 的bounding box也没有。但是是有测试结果的。 请问这是为什么呢？ 我的数据集结构： --data_usecase3_yoloworld --train --val --annotations --val.json --train.json 用mmyolo的analysis_tools, browse_coco_json.py是能显示出GT的bounding box的。后面我用image_demo.py测试也是能有bounding box的。 所以我不太清楚是哪里错了。谢谢！

wondervictor commented 6 months ago

Hi @mandyxiaomeng, you could add a --show-dir with your targeted directory. BTW, the visualization has a threshold.

mandyxiaomeng commented 6 months ago

Hi @mandyxiaomeng, you could add a --show-dir with your targeted directory. BTW, the visualization has a threshold.

Hi @wondervictor , Thank you for the reply. I found out it is not a problem of visualization but a problem of training. My trained model cannot detect anything.

The problem is:

  1. loss_bbox and loss_dfl from the beginning is 0.
  2. During evaluation, it says 'The testing results of the whole dataset are empty', and the trained model doesn't give any results while predicting new images.

This is exactly like #146 , #126 and #185. I have tried different methods mentioned in these three topics, but still cannot solve the problem. I think my metainfo and load_from are correct? Do you have any suggestions? Thank you!

Below is the detailed info I got during training. 

05/08 17:16:01 - mmengine - INFO - Epoch(train) [1][ 50/655]  base_lr: 2.0000e-05 lr: 4.9873e-07  eta: 0:13:08  time: 1.3025  data_time: 0.8345  memory: 34153  grad_norm: nan  loss: 609.7719  loss_cls: 609.7719  loss_bbox: 0.0000  loss_dfl: 0.0000
05/08 17:17:01 - mmengine - INFO - Epoch(train) [1][100/655]  base_lr: 2.0000e-05 lr: 1.0076e-06  eta: 0:11:34  time: 1.1991  data_time: 0.9747  memory: 10640  grad_norm: 27732.7017  loss: 442.9217  loss_cls: 442.9217  loss_bbox: 0.0000  loss_dfl: 0.0000
05/08 17:17:58 - mmengine - INFO - Epoch(train) [1][150/655]  base_lr: 2.0000e-05 lr: 1.5165e-06  eta: 0:10:11  time: 1.1300  data_time: 0.9082  memory: 10640  grad_norm: 16402.5593  loss: 255.1138  loss_cls: 255.1138  loss_bbox: 0.0000  loss_dfl: 0.0000
05/08 17:18:52 - mmengine - INFO - Epoch(train) [1][200/655]  base_lr: 2.0000e-05 lr: 2.0254e-06  eta: 0:08:57  time: 1.0905  data_time: 0.8680  memory: 10640  grad_norm: 11682.1961  loss: 128.2933  loss_cls: 128.2933  loss_bbox: 0.0000  loss_dfl: 0.0000
05/08 17:19:48 - mmengine - INFO - Epoch(train) [1][250/655]  base_lr: 2.0000e-05 lr: 2.5344e-06  eta: 0:07:52  time: 1.1172  data_time: 0.8876  memory: 10640  grad_norm: 6005.4033  loss: 60.0497  loss_cls: 60.0497  loss_bbox: 0.0000  loss_dfl: 0.0000
05/08 17:20:45 - mmengine - INFO - Epoch(train) [1][300/655]  base_lr: 2.0000e-05 lr: 3.0433e-06  eta: 0:06:52  time: 1.1387  data_time: 0.9109  memory: 10640  grad_norm: 2919.8520  loss: 24.3001  loss_cls: 24.3001  loss_bbox: 0.0000  loss_dfl: 0.0000
05/08 17:21:44 - mmengine - INFO - Epoch(train) [1][350/655]  base_lr: 2.0000e-05 lr: 3.5522e-06  eta: 0:05:55  time: 1.1786  data_time: 0.9514  memory: 10640  grad_norm: 797.1709  loss: 7.0634  loss_cls: 7.0634  loss_bbox: 0.0000  loss_dfl: 0.0000
05/08 17:22:43 - mmengine - INFO - Epoch(train) [1][400/655]  base_lr: 2.0000e-05 lr: 4.0611e-06  eta: 0:04:57  time: 1.1869  data_time: 0.9600  memory: 10640  grad_norm: 553.4262  loss: 2.1420  loss_cls: 2.1420  loss_bbox: 0.0000  loss_dfl: 0.0000
05/08 17:23:34 - mmengine - INFO - Epoch(train) [1][450/655]  base_lr: 2.0000e-05 lr: 4.5700e-06  eta: 0:03:55  time: 1.0127  data_time: 0.7891  memory: 10640  grad_norm: 260.8789  loss: 0.5537  loss_cls: 0.5537  loss_bbox: 0.0000  loss_dfl: 0.0000
05/08 17:24:27 - mmengine - INFO - Epoch(train) [1][500/655]  base_lr: 2.0000e-05 lr: 5.0789e-06  eta: 0:02:56  time: 1.0512  data_time: 0.8300  memory: 10640  grad_norm: 47.5409  loss: 0.0902  loss_cls: 0.0902  loss_bbox: 0.0000  loss_dfl: 0.0000
05/08 17:25:26 - mmengine - INFO - Epoch(train) [1][550/655]  base_lr: 2.0000e-05 lr: 5.5878e-06  eta: 0:02:00  time: 1.1815  data_time: 0.9594  memory: 10640  grad_norm: 9.8181  loss: 0.0138  loss_cls: 0.0138  loss_bbox: 0.0000  loss_dfl: 0.0000
05/08 17:26:22 - mmengine - INFO - Epoch(train) [1][600/655]  base_lr: 2.0000e-05 lr: 6.0967e-06  eta: 0:01:02  time: 1.1224  data_time: 0.8993  memory: 10640  grad_norm: 22.4686  loss: 0.0083  loss_cls: 0.0083  loss_bbox: 0.0000  loss_dfl: 0.0000
05/08 17:27:13 - mmengine - INFO - Epoch(train) [1][650/655]  base_lr: 2.0000e-05 lr: 6.6056e-06  eta: 0:00:05  time: 1.0293  data_time: 0.8067  memory: 10640  grad_norm: 2.7762  loss: 0.0034  loss_cls: 0.0034  loss_bbox: 0.0000  loss_dfl: 0.0000
05/08 17:27:15 - mmengine - INFO - Exp name: yolo_world_l_usecase3_random2_20240508_171439
05/08 17:27:15 - mmengine - INFO - Saving checkpoint at 1 epochs
05/08 17:27:18 - mmengine - WARNING - `save_param_scheduler` is True but `self.param_schedulers` is None, so skip saving parameter schedulers
huggingface/tokenizers: The current process just got forked, after parallelism has already been used. Disabling parallelism to avoid deadlocks...
To disable this warning, you can either:
    - Avoid using `tokenizers` before the fork if possible
    - Explicitly set the environment variable TOKENIZERS_PARALLELISM=(true | false)
huggingface/tokenizers: The current process just got forked, after parallelism has already been used. Disabling parallelism to avoid deadlocks...
To disable this warning, you can either:
    - Avoid using `tokenizers` before the fork if possible
    - Explicitly set the environment variable TOKENIZERS_PARALLELISM=(true | false)
05/08 17:27:23 - mmengine - INFO - Epoch(val) [1][  50/2621]    eta: 0:02:07  time: 0.0496  data_time: 0.0039  memory: 10640  
05/08 17:27:25 - mmengine - INFO - Epoch(val) [1][ 100/2621]    eta: 0:01:43  time: 0.0327  data_time: 0.0003  memory: 1638  
05/08 17:27:26 - mmengine - INFO - Epoch(val) [1][ 150/2621]    eta: 0:01:34  time: 0.0322  data_time: 0.0003  memory: 1638  
05/08 17:27:28 - mmengine - INFO - Epoch(val) [1][ 200/2621]    eta: 0:01:28  time: 0.0321  data_time: 0.0003  memory: 1638  
05/08 17:27:30 - mmengine - INFO - Epoch(val) [1][ 250/2621]    eta: 0:01:24  time: 0.0321  data_time: 0.0003  memory: 1638  
05/08 17:27:31 - mmengine - INFO - Epoch(val) [1][ 300/2621]    eta: 0:01:21  time: 0.0314  data_time: 0.0003  memory: 1638  
05/08 17:27:33 - mmengine - INFO - Epoch(val) [1][ 350/2621]    eta: 0:01:18  time: 0.0323  data_time: 0.0003  memory: 1638  
05/08 17:27:34 - mmengine - INFO - Epoch(val) [1][ 400/2621]    eta: 0:01:16  time: 0.0329  data_time: 0.0003  memory: 1638  
05/08 17:27:36 - mmengine - INFO - Epoch(val) [1][ 450/2621]    eta: 0:01:15  time: 0.0359  data_time: 0.0004  memory: 1638  
05/08 17:27:38 - mmengine - INFO - Epoch(val) [1][ 500/2621]    eta: 0:01:13  time: 0.0339  data_time: 0.0004  memory: 1638  
05/08 17:27:40 - mmengine - INFO - Epoch(val) [1][ 550/2621]    eta: 0:01:11  time: 0.0334  data_time: 0.0003  memory: 1638  
05/08 17:27:41 - mmengine - INFO - Epoch(val) [1][ 600/2621]    eta: 0:01:09  time: 0.0333  data_time: 0.0004  memory: 1638  
05/08 17:27:43 - mmengine - INFO - Epoch(val) [1][ 650/2621]    eta: 0:01:07  time: 0.0328  data_time: 0.0003  memory: 1638  
05/08 17:27:45 - mmengine - INFO - Epoch(val) [1][ 700/2621]    eta: 0:01:05  time: 0.0354  data_time: 0.0004  memory: 1638  
05/08 17:27:46 - mmengine - INFO - Epoch(val) [1][ 750/2621]    eta: 0:01:04  time: 0.0335  data_time: 0.0003  memory: 1638  
05/08 17:27:48 - mmengine - INFO - Epoch(val) [1][ 800/2621]    eta: 0:01:02  time: 0.0337  data_time: 0.0004  memory: 1638  
05/08 17:27:50 - mmengine - INFO - Epoch(val) [1][ 850/2621]    eta: 0:01:00  time: 0.0354  data_time: 0.0004  memory: 1638  
05/08 17:27:52 - mmengine - INFO - Epoch(val) [1][ 900/2621]    eta: 0:00:58  time: 0.0345  data_time: 0.0003  memory: 1638  
05/08 17:27:53 - mmengine - INFO - Epoch(val) [1][ 950/2621]    eta: 0:00:57  time: 0.0350  data_time: 0.0003  memory: 1638  
05/08 17:27:55 - mmengine - INFO - Epoch(val) [1][1000/2621]    eta: 0:00:55  time: 0.0337  data_time: 0.0004  memory: 1638  
05/08 17:27:57 - mmengine - INFO - Epoch(val) [1][1050/2621]    eta: 0:00:53  time: 0.0357  data_time: 0.0004  memory: 1638  
05/08 17:27:59 - mmengine - INFO - Epoch(val) [1][1100/2621]    eta: 0:00:52  time: 0.0340  data_time: 0.0003  memory: 1638  
05/08 17:28:00 - mmengine - INFO - Epoch(val) [1][1150/2621]    eta: 0:00:50  time: 0.0350  data_time: 0.0003  memory: 1638  
05/08 17:28:02 - mmengine - INFO - Epoch(val) [1][1200/2621]    eta: 0:00:48  time: 0.0351  data_time: 0.0003  memory: 1638  
05/08 17:28:04 - mmengine - INFO - Epoch(val) [1][1250/2621]    eta: 0:00:47  time: 0.0353  data_time: 0.0003  memory: 1638  
05/08 17:28:05 - mmengine - INFO - Epoch(val) [1][1300/2621]    eta: 0:00:45  time: 0.0325  data_time: 0.0003  memory: 1638  
05/08 17:28:07 - mmengine - INFO - Epoch(val) [1][1350/2621]    eta: 0:00:43  time: 0.0345  data_time: 0.0003  memory: 1638  
05/08 17:28:09 - mmengine - INFO - Epoch(val) [1][1400/2621]    eta: 0:00:42  time: 0.0359  data_time: 0.0004  memory: 1638  
05/08 17:28:11 - mmengine - INFO - Epoch(val) [1][1450/2621]    eta: 0:00:40  time: 0.0364  data_time: 0.0003  memory: 1638  
05/08 17:28:13 - mmengine - INFO - Epoch(val) [1][1500/2621]    eta: 0:00:38  time: 0.0361  data_time: 0.0003  memory: 1638  
05/08 17:28:14 - mmengine - INFO - Epoch(val) [1][1550/2621]    eta: 0:00:36  time: 0.0339  data_time: 0.0003  memory: 1638  
05/08 17:28:16 - mmengine - INFO - Epoch(val) [1][1600/2621]    eta: 0:00:35  time: 0.0351  data_time: 0.0003  memory: 1638  
05/08 17:28:18 - mmengine - INFO - Epoch(val) [1][1650/2621]    eta: 0:00:33  time: 0.0340  data_time: 0.0003  memory: 1638  
05/08 17:28:19 - mmengine - INFO - Epoch(val) [1][1700/2621]    eta: 0:00:31  time: 0.0349  data_time: 0.0003  memory: 1638  
05/08 17:28:21 - mmengine - INFO - Epoch(val) [1][1750/2621]    eta: 0:00:30  time: 0.0335  data_time: 0.0003  memory: 1638  
05/08 17:28:23 - mmengine - INFO - Epoch(val) [1][1800/2621]    eta: 0:00:28  time: 0.0336  data_time: 0.0003  memory: 1638  
05/08 17:28:25 - mmengine - INFO - Epoch(val) [1][1850/2621]    eta: 0:00:26  time: 0.0347  data_time: 0.0003  memory: 1638  
05/08 17:28:26 - mmengine - INFO - Epoch(val) [1][1900/2621]    eta: 0:00:24  time: 0.0345  data_time: 0.0004  memory: 1638  
05/08 17:28:28 - mmengine - INFO - Epoch(val) [1][1950/2621]    eta: 0:00:23  time: 0.0333  data_time: 0.0003  memory: 1638  
05/08 17:28:30 - mmengine - INFO - Epoch(val) [1][2000/2621]    eta: 0:00:21  time: 0.0340  data_time: 0.0003  memory: 1638  
05/08 17:28:31 - mmengine - INFO - Epoch(val) [1][2050/2621]    eta: 0:00:19  time: 0.0343  data_time: 0.0003  memory: 1638  
05/08 17:28:33 - mmengine - INFO - Epoch(val) [1][2100/2621]    eta: 0:00:17  time: 0.0329  data_time: 0.0003  memory: 1638  
05/08 17:28:35 - mmengine - INFO - Epoch(val) [1][2150/2621]    eta: 0:00:16  time: 0.0312  data_time: 0.0003  memory: 1638  
05/08 17:28:36 - mmengine - INFO - Epoch(val) [1][2200/2621]    eta: 0:00:14  time: 0.0318  data_time: 0.0003  memory: 1638  
05/08 17:28:38 - mmengine - INFO - Epoch(val) [1][2250/2621]    eta: 0:00:12  time: 0.0332  data_time: 0.0003  memory: 1638  
05/08 17:28:40 - mmengine - INFO - Epoch(val) [1][2300/2621]    eta: 0:00:10  time: 0.0345  data_time: 0.0003  memory: 1638  
05/08 17:28:41 - mmengine - INFO - Epoch(val) [1][2350/2621]    eta: 0:00:09  time: 0.0343  data_time: 0.0003  memory: 1638  
05/08 17:28:43 - mmengine - INFO - Epoch(val) [1][2400/2621]    eta: 0:00:07  time: 0.0323  data_time: 0.0003  memory: 1638  
05/08 17:28:45 - mmengine - INFO - Epoch(val) [1][2450/2621]    eta: 0:00:05  time: 0.0334  data_time: 0.0003  memory: 1638  
05/08 17:28:46 - mmengine - INFO - Epoch(val) [1][2500/2621]    eta: 0:00:04  time: 0.0340  data_time: 0.0003  memory: 1638  
05/08 17:28:48 - mmengine - INFO - Epoch(val) [1][2550/2621]    eta: 0:00:02  time: 0.0339  data_time: 0.0003  memory: 1638  
05/08 17:28:50 - mmengine - INFO - Epoch(val) [1][2600/2621]    eta: 0:00:00  time: 0.0340  data_time: 0.0003  memory: 1638  
05/08 17:28:50 - mmengine - INFO - Evaluating bbox...
Loading and preparing results...
05/08 17:28:50 - mmengine - ERROR - /proj/simtoreal_mandy/users/x_xizhu/Project/Paper1/mmdetection-3.0.0/mmdet/evaluation/metrics/coco_metric.py - compute_metrics - 461 - The testing results of the whole dataset is empty.
05/08 17:28:50 - mmengine - INFO - Epoch(val) [1][2621/2621]    data_time: 0.0004  time: 0.0342
05/08 17:28:50 - mmengine - WARNING - Since `metrics` is an empty dict, the behavior to save the best checkpoint will be skipped in this evaluation.
LRuiRui517 commented 3 months ago

I have the same problem. It's the same. Have you solved it?

LLH-Harward commented 3 months ago

@LRuiRui517 https://github.com/AILab-CVC/YOLO-World/issues/126