FishmanTheMurloc
4787f4f2f9
尝试复现RodNet的训练结果 (#1688)
* 增加RodNet必须支持训练的单元测试;RodNet的torch链路改造成全张量计算,由此把之前忽略的参数dz、h_coeff、offset变得可学习
* 了解到损失函数CrossEntropyLoss内置了softmax,因此从forward方法中移出;offset是手动指定的偏置值,不是学习得到的,因此也移出到单独的PostProcess方法中
* 根据得到的源码整理RodNet;新增在数据集上达到一定准确率的单元测试
2025-06-14 20:03:14 +08:00
..
2025-04-05 19:53:52 +08:00
2024-05-26 19:14:44 +08:00
2025-06-02 00:21:01 +08:00
2025-06-08 17:47:11 +08:00
2025-06-14 20:03:14 +08:00
2025-05-12 18:05:15 +08:00
2025-02-04 17:11:53 +08:00
2025-06-07 22:13:57 +08:00
2025-05-11 11:03:53 +08:00
2025-05-11 11:03:53 +08:00
2025-06-07 22:13:57 +08:00
2025-05-11 01:08:37 +08:00
2025-06-14 10:49:23 +08:00
2025-02-22 18:56:25 +08:00
2025-04-26 21:47:57 +08:00
2024-08-20 23:46:29 +08:00
2025-05-11 11:03:53 +08:00
2025-05-11 11:03:53 +08:00
2024-08-20 23:46:29 +08:00
2024-08-14 18:01:02 +08:00
2025-01-06 22:55:08 +08:00
2025-05-11 11:03:53 +08:00
2025-03-15 20:35:53 +08:00
2025-06-02 00:03:26 +08:00
2024-08-14 18:01:02 +08:00
2025-05-11 10:48:47 +08:00
2025-04-26 17:35:21 +08:00
2024-10-19 17:42:58 +08:00
2024-06-01 18:09:34 +08:00
2025-06-02 00:03:26 +08:00
2025-06-02 00:03:26 +08:00
2025-04-02 21:40:14 +08:00
2025-03-23 22:04:43 +08:00
2025-05-11 01:17:18 +08:00