search

SwinTransformer / Swin-Transformer-Object-Detection

This is an official implementation for "Swin Transformer: Hierarchical Vision Transformer using Shifted Windows" on Object Detection and Instance Segmentation.
https://arxiv.org/abs/2103.14030
Apache License 2.0
1.76k stars 375 forks source link

Transformer Swin backbone to other Detector get 0 AP #55

Open Kingdroper opened 3 years ago

Kingdroper commented 3 years ago

model = dict( type='FCOS', pretrained=None, backbone=dict( type='SwinTransformer', embed_dim=96, depths=[2, 2, 6, 2], num_heads=[3, 6, 12, 24], window_size=7, mlp_ratio=4., qkv_bias=True, qk_scale=None, drop_rate=0., attn_drop_rate=0., drop_path_rate=0.2, ape=False, patch_norm=True, out_indices=(0, 1, 2, 3), use_checkpoint=False), neck=dict( type='FPN', in_channels=[96, 192, 384, 768], out_channels=256, num_outs=5), bbox_head=dict( type='FCOSHead', num_classes=11, in_channels=256, stacked_convs=4, feat_channels=256, strides=[8, 16, 32, 64, 128], loss_cls=dict( type='FocalLoss', use_sigmoid=True, gamma=2.0, alpha=0.25, loss_weight=1.0), loss_bbox=dict(type='IoULoss', loss_weight=1.0), loss_centerness=dict( type='CrossEntropyLoss', use_sigmoid=True, loss_weight=1.0)), # For example, I Ctrl+C the configs of Swin backbone and FPN from cascade_mask_rcnn_swin_tiny_patch4_window7_mstrain_480-800_giou_4conv1f_adamw_1x_coco.py, And Ctrl+V to Fcos Detector. I use the changed Swin-Fcos Configs to train a detector and get 0 AP. What is wrong with it? Could you give me a hand?

SISTMrL commented 3 years ago

I replace the backbone of detr with swin, i also get 0 ap， do you solve this problem? thanks!

WassimBouatay commented 2 years ago

I also tried Swin with YoloV3 but i also get 0 ap

sfx6635 commented 2 years ago

Have you solved this problem？

我用swin替换了detr的主干，我也得到了0 ap，你解决了这个问题吗？谢谢！