mcmingchang
mcmingchang
框的置信度有0.5到0.8,但是mask就是没有,是不是segment的值太小了
![123](https://user-images.githubusercontent.com/37234299/178463577-68c7cbfe-7ff2-4bea-8496-93280cf8c1e7.png)
相同的问题,在一张卡上能跑,问题出现在yolox/core/launch.py 的mp.start_processes,偶尔2张卡也能跑
what about UNeXt? https://github.com/jeya-maria-jose/UNeXt-pytorch
My data dataAugment has greatly improved my test results, better than cutting def down_sample_drop(self, xyz, rgb, semantic_label, instance_label, drop_prob=0.1, min_voxel=0.01, max_voxel=0.03): xyz_ls, rgb_ls, selabel_ls = [], [], [] for i...
in my custom dataset, mIoU from 85.3 to 91.8, acc from 96.5 to 98.2. mlp dense num layers 3 act=nn.silu new dataAugment down_sample_drop spatial_shape: [256, 1024]
I built mobilenetv2 v3 in the same way, they all work well, I can provide the codes of resnest and vovnet to you.
I tried to use dlav0_34 to train the birds, dogs, and cats of the coco dataset. The first training was very successful, but the second time Loss got stuck at...
https://github.com/CaoWGG/Mobilenetv2-CenterNet
export onnx successful tgt_padding_mask = (((tgt_in == self.eos_id)*2).cumsum(-1) > 0) # mask tokens beyond the first EOS token.