Lei-Lei Ma
Lei-Lei Ma
> I also get NaNs when using ASL on a custom multi-label classification task. Everything seemed to work fine when I tested with gamma_neg=0, gamma_pos=0 and gamma_neg=2, gamma_pos=2. However, it...
> I also had a hard time and mine is 83.1 using resnet101 ,and I followed the tricks in [#30 (comment)](https://github.com/Alibaba-MIIL/ASL/issues/30#issuecomment-750780576) Would you like to provide a training log?
> mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_rgb=True)??? 1: Imagenet Normalize mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225] 2: cutout The server where the previous training logs are located is currently...
@THUeeY Are you interested in adding your email address to your homepage? Emails can be sent between us. [07/13 17:40:44.800]: Epoch: [0/80][ 0/646] T 8.995 (8.995) DT 2.315 (2.315) S1...
ASL's convergence speed is more excellent than BCE's, and the final mAP is almost the same.
```python available_models = ['Q2L-R101-448', 'Q2L-R101-576', 'Q2L-TResL-448', 'Q2L-TResL_22k-448', 'Q2L-SwinL-384', 'Q2L-CvT_w24-384'] parser = argparse.ArgumentParser(description='Query2Label MSCOCO Training') parser.add_argument('--dataname', help='dataname', default='coco14', choices=['coco14']) parser.add_argument('--dataset_dir', help='dir of dataset', default='/comp_robot/liushilong/data/COCO14/') parser.add_argument('--img_size', default=448, type=int, help='size of input images')...
RuntimeError: The size of tensor a (1024) must match the size of tensor b (2048) at non-singleton dimension 2
你对pos_embedding做多个实例化 或者你试试这个 `PositionEmbeddingLearned(nn.Module):` https://github.com/facebookresearch/detr/blob/main/models/position_encoding.py
> > > > @zhgqcn 静态图的参数是通过网络学习获得的，在训练完毕之后，静态图的参数就 fix 了。即在测试阶段：**静态图是对所属输入数据共享的** > > > > > > > > > 在您的代码中没用看到Fix操作，好像保存和加载的都是整个模型的参数？可以提供完整的代码参考吗，非常感谢 > > > > > > 这里 fix 的意思就是模型参数啊，模型参数在训练完成之后就固定了不会改变了啊 :) > >...
This trick is mainly used for object detection with very small batches-size. Nevertheless, the batch size in this method is relatively large, so I suggest you remove FrozenBatchNorm2d.