Are you sure you want to delete this task? Once this task is deleted, it cannot be recovered.
zrainj 553093b5d1 | 7 months ago | |
---|---|---|
data | 1 year ago | |
save_model | 1 year ago | |
src | 1 year ago | |
utils | 1 year ago | |
.gitignore | 1 year ago | |
LICENSE | 1 year ago | |
README.md | 1 year ago | |
alpha1-10-01.log | 7 months ago | |
eval.py | 1 year ago | |
requirements.txt | 1 year ago | |
train.py | 1 year ago |
模型: ConvLSTM
论文链接: https://arxiv.org/pdf/1506.04214.pdf
简介:
模型ConvLSTM作者通过实验**证明了ConvLSTM在获取时空关系上比LSTM有更好的效果。**ConvLSTM不仅可以预测天气,还能够解决其他时空序列的预测问题。比如视频分类,动作识别等。此次数据集为Moving MNIST
训练数据集: The MNIST database of handwritten digits
验证精度数据集: Moving MNIST
下载数据集请通过目录下的脚本下载
source ./download.sh
Environment(Ascend/GPU/CPU): GPU-GTX3090(24G)
Software Environment:
– MindSpore version : 1.7.0
– Python version : 3.8.13
– OS platform and distribution : Ubuntu 16.04
– CUDA version : 11.0
训练命令:
python train.py --batch_size 24 -checkpoints 'checkpoint_66_0.000961.ckpt' -epochs 500
评估命令:
python eval.py --batch_size 24 -checkpoints 'checkpoint_66_0.000961.ckpt'
训练过程采用MNIST手写数字数据库,其中有60,000个示例的训练集和10,000个示例的测验集。它是MNIST的子集。数字已被归一化并以固定大小的图像为中心。训练及测验过程中通过动态生成视频数据来进行训练。特别需要注意,训练过程中生成的数字数量为3,相较于评估中的2个数字更多。
通过以下指令启动训练。保存的参数模型将存于当前目录save_model中
python train.py --batch_size 24 -checkpoints 'checkpoint_66_0.000961.ckpt' -epochs 500
评估过程采用Moving MNIST作为测试集。
通过以下指令启动评估。
python eval.py --batch_size 24 -checkpoints 'checkpoint_66_0.000961.ckpt'
train_loss | valid_loss | SSIM | MAE | MSE |
---|---|---|---|---|
0.000976 | 0.000961 | 0.777687 | 221.285598 | 94.498799 |
test_loss | SSIM | MAE | MSE |
---|---|---|---|
0.000638 | 0.833904 | 156.482312 | 62.759463 |
载入权重模型后继续训练会有较大精度浮动
请浏览官方主页。
Dear OpenI User
Thank you for your continuous support to the Openl Qizhi Community AI Collaboration Platform. In order to protect your usage rights and ensure network security, we updated the Openl Qizhi Community AI Collaboration Platform Usage Agreement in January 2024. The updated agreement specifies that users are prohibited from using intranet penetration tools. After you click "Agree and continue", you can continue to use our services. Thank you for your cooperation and understanding.
For more agreement content, please refer to the《Openl Qizhi Community AI Collaboration Platform Usage Agreement》