PANPEIWEN / ABC

ABC: Attention with Bilinear Correlation for Infrared Small Target Detection
MIT License
26 stars 2 forks source link

模型权重文件 #5

Open orange998 opened 1 year ago

orange998 commented 1 year ago

你好,读了你的论文后,我受到了一些启发,但是在做实验复现时结果有点不太理想,可以提供下模型网络关于每个数据集的权重文件嘛,best.pth.tar

PANPEIWEN commented 1 year ago

你好 你能把你的训练细节说一下嘛 我这两天就会把训练好的模型权重整理好

orange998 @.***> 于2023年4月13日周四 20:26写道:

你好,读了你的论文后,我受到了一些启发,但是在做实验复现时结果有点不太理想,可以提供下模型网络关于每个数据集的权重文件嘛,best.pth.tar

— Reply to this email directly, view it on GitHub https://github.com/PANPEIWEN/ABC/issues/5, or unsubscribe https://github.com/notifications/unsubscribe-auth/AWNTCO34QTMZDZ5BNKM42BLXA7WHLANCNFSM6AAAAAAW5AHELM . You are receiving this because you are subscribed to this thread.Message ID: @.***>

orange998 commented 1 year ago

data = dict(base_size=256, crop_size=256, data_aug=True, suffix='png', num_workers=8, train_batch=32, test_batch=16, train_dir='trainval', test_dir='test') log_config = dict( interval=10, hooks=[ dict(type='TextLoggerHook', by_epoch=True), dict(type='TensorboardLoggerHook') ]) dist_params = dict(backend='nccl') log_level = 'INFO' load_from = None resume_from = None workflow = [('train', 1)] cudnn_benchmark = True gpus = 1 optimizer = dict( type='AdamW', setting=dict(lr=0.0001, weight_decay=0.01, betas=(0.9, 0.999))) optimizer_config = dict() lr_config = dict( policy='PolyLR', warmup='linear', power=0.9, min_lr=0.0001, warmup_epochs=5) runner = dict(type='EpochBasedRunner', max_epochs=1500) checkpoint_config = dict(by_epoch=False, interval=1) evaluation = dict(epochval=1, metric='mIoU', pre_eval=True) model = dict( name='ABCNet', type='EncoderDecoder', pretrained=None, backbone=dict(type=None), decode_head=dict( type='ABCNet', in_ch=3, out_ch=1, dim=16, ori_h=256, deep_supervision=True), loss=dict(type='SoftIoULoss')) 我没有对默认文件做改动,只更改了epochs,这是训练nudt数据集的信息

PANPEIWEN commented 1 year ago

你好 你使用的是abcnet_clft-s_256x256_500e_nudt.py https://github.com/PANPEIWEN/ABC/blob/main/configs/abcnet/abcnet_clft-s_256x256_1500e_nudt.py 这个文件对应的模型最小 性能也是最差的 我们的最好结果是用abcnet_clft-l_256x256_500e_nudt.py https://github.com/PANPEIWEN/ABC/blob/main/configs/abcnet/abcnet_clft-l_256x256_500e_nudt.py 这个文件训练出来的 另外 这个数据集收敛的较慢 并且我们的学习率衰减策略是根据epochs来决定的 所以建议你保持原来的epochs epochs默认为1500

orange998 @.***> 于2023年4月15日周六 15:24写道:

data = dict(base_size=256, crop_size=256, data_aug=True, suffix='png', num_workers=8, train_batch=32, test_batch=16, train_dir='trainval', test_dir='test') log_config = dict( interval=10, hooks=[ dict(type='TextLoggerHook', by_epoch=True), dict(type='TensorboardLoggerHook') ]) dist_params = dict(backend='nccl') log_level = 'INFO' load_from = None resume_from = None workflow = [('train', 1)] cudnn_benchmark = True gpus = 1 optimizer = dict( type='AdamW', setting=dict(lr=0.0001, weight_decay=0.01, betas=(0.9, 0.999))) optimizer_config = dict() lr_config = dict( policy='PolyLR', warmup='linear', power=0.9, min_lr=0.0001, warmup_epochs=5) runner = dict(type='EpochBasedRunner', max_epochs=1500) checkpoint_config = dict(by_epoch=False, interval=1) evaluation = dict(epochval=1, metric='mIoU', pre_eval=True) model = dict( name='ABCNet', type='EncoderDecoder', pretrained=None, backbone=dict(type=None), decode_head=dict( type='ABCNet', in_ch=3, out_ch=1, dim=16, ori_h=256, deep_supervision=True), loss=dict(type='SoftIoULoss')) 我没有对默认文件做改动,只更改了epochs,这是训练nudt数据集的信息

— Reply to this email directly, view it on GitHub https://github.com/PANPEIWEN/ABC/issues/5#issuecomment-1509608574, or unsubscribe https://github.com/notifications/unsubscribe-auth/AWNTCO74WT6FIWNB27FOIRLXBJELBANCNFSM6AAAAAAW5AHELM . You are receiving this because you commented.Message ID: @.***>

orange998 commented 1 year ago

好的,十分感谢

RenYong1 commented 1 year ago

楼主,可以提供一下你的环境配置相关包的信息吗?我是windows系统,代码调了很久都调不通

orange998 commented 1 year ago

你可以在云服务器上试一下环境,我记得我的cuda是11.3,pytorch==1.12.0 torchvision==0.13.0 torchaudio==0.12.0 cudatoolkit=11.3,其他的缺什么直接装的

RenYong1 commented 1 year ago

好的,十分感谢,再请教一下你是linux系统还是windows系统捏?

PANPEIWEN commented 1 year ago

好的,十分感谢,再请教一下你是linux系统还是windows系统捏?

是Linux教程

RenYong1 commented 1 year ago

好的,感谢指教