PaddleX
PaddleX copied to clipboard
PaddlePaddle
模型训练Faster RCNN时出现IndexError: list index out of range
Checklist:
- 查找历史相关issue寻求解答
- 翻阅FAQ常见问题汇总和答疑
- 确认bug是否在新版本里还未修复
- 翻阅PaddleX API文档说明
描述问题
当我在Al studio尝试训练自制的标注集来测试时，在模型训练的过程中会出现IndexError: list index out of range的报错.但奇怪的是就在发布此issue的前一晚，是可以正常的进行模型训练，期间没改变过任何环境配置。重新创建新项目复现此环境依然会发生此错误。
复现
-
您是否已经正常运行我们提供的教程？ 是
-
您是否在教程的基础上修改代码内容？还请您提供运行的代码
!cd ~ # 设置当前目录路径 !unzip data/data132692/VOCdataset.zip -d work/dataset # 将自制数据集进行解压 !pip install paddlex==2.1.0 -i https://mirror.baidu.com/pypi/simple # 下载paddlex
!paddlex --split_dataset --format VOC --dataset_dir 'work/dataset/VOCdataset' --val_value 0.2 --test_value 0.1
数据增强
import paddlex as pdx from paddlex import transforms as T train_transforms = T.Compose([ T.RandomHorizontalFlip(), T.RandomDistort(), T.RandomExpand(), T.RandomCrop(), T.ResizeByShort(short_size=250, max_size=-1), T.Normalize() ]) eval_transforms = T.Compose([ T.ResizeByShort(short_size=250, max_size=-1), T.Normalize() ])
加载数据集
train_dataset = pdx.datasets.VOCDetection( data_dir=r'work/dataset/VOCdataset', file_list=r'work/dataset/VOCdataset/train_list.txt', label_list=r'work/dataset/VOCdataset/labels.txt', transforms=train_transforms, # shuffle=True ) eval_dataset = pdx.datasets.VOCDetection( data_dir=r'work/dataset/VOCdataset', file_list=r'work/dataset/VOCdataset/val_list.txt', label_list=r'work/dataset/VOCdataset/labels.txt', transforms=eval_transforms)
模型训练 Faster RCNN
num_classes = len(train_dataset.labels)+1 print(num_classes) model = pdx.det.FasterRCNN(num_classes=num_classes, backbone='ResNet50') model.train( num_epochs=10, train_dataset=train_dataset, train_batch_size=1, eval_dataset=eval_dataset, save_interval_epochs=1, learning_rate=0.00025, # learning_rate=0.0001, # lr_decay_epochs = [3,6], # lr_decay_gamma = 0.1, metric='VOC', save_dir='output/faster_rcnn_r50_fpn', use_vdl=True)
至此模型训练出现 IndexError: list index out of range，报错信息在BUG描述中。
-
您使用的数据集是？ （自己使用labelImg标注的10张图片，用于测试） VOCdataset.zip
-
请提供您出现的报错信息及相关log
执行代码：
num_classes = len(train_dataset.labels)+1 print(num_classes) model = pdx.det.FasterRCNN(num_classes=num_classes, backbone='ResNet50') model.train( num_epochs=10, train_dataset=train_dataset, train_batch_size=1, eval_dataset=eval_dataset, save_interval_epochs=1, learning_rate=0.00025, metric='VOC', save_dir='output/faster_rcnn_r50_fpn', use_vdl=True)
BUG描述：
2023-03-08 16:08:03 [INFO] Loading pretrained model from output/faster_rcnn_r50_fpn/pretrain/ResNet50_cos_pretrained.pdparams
2023-03-08 16:08:04 [WARNING] neck.fpn_inner_res2_sum_lateral.weight is not in pretrained model
2023-03-08 16:08:04 [WARNING] neck.fpn_inner_res2_sum_lateral.bias is not in pretrained model
2023-03-08 16:08:04 [WARNING] neck.fpn_res2_sum.weight is not in pretrained model
2023-03-08 16:08:04 [WARNING] neck.fpn_res2_sum.bias is not in pretrained model
2023-03-08 16:08:04 [WARNING] neck.fpn_inner_res3_sum_lateral.weight is not in pretrained model
2023-03-08 16:08:04 [WARNING] neck.fpn_inner_res3_sum_lateral.bias is not in pretrained model
2023-03-08 16:08:04 [WARNING] neck.fpn_res3_sum.weight is not in pretrained model
2023-03-08 16:08:04 [WARNING] neck.fpn_res3_sum.bias is not in pretrained model
2023-03-08 16:08:04 [WARNING] neck.fpn_inner_res4_sum_lateral.weight is not in pretrained model
2023-03-08 16:08:04 [WARNING] neck.fpn_inner_res4_sum_lateral.bias is not in pretrained model
2023-03-08 16:08:04 [WARNING] neck.fpn_res4_sum.weight is not in pretrained model
2023-03-08 16:08:04 [WARNING] neck.fpn_res4_sum.bias is not in pretrained model
2023-03-08 16:08:04 [WARNING] neck.fpn_inner_res5_sum.weight is not in pretrained model
2023-03-08 16:08:04 [WARNING] neck.fpn_inner_res5_sum.bias is not in pretrained model
2023-03-08 16:08:04 [WARNING] neck.fpn_res5_sum.weight is not in pretrained model
2023-03-08 16:08:04 [WARNING] neck.fpn_res5_sum.bias is not in pretrained model
2023-03-08 16:08:04 [WARNING] rpn_head.rpn_feat.rpn_conv.weight is not in pretrained model
2023-03-08 16:08:04 [WARNING] rpn_head.rpn_feat.rpn_conv.bias is not in pretrained model
2023-03-08 16:08:04 [WARNING] rpn_head.rpn_rois_score.weight is not in pretrained model
2023-03-08 16:08:04 [WARNING] rpn_head.rpn_rois_score.bias is not in pretrained model
2023-03-08 16:08:04 [WARNING] rpn_head.rpn_rois_delta.weight is not in pretrained model
2023-03-08 16:08:04 [WARNING] rpn_head.rpn_rois_delta.bias is not in pretrained model
2023-03-08 16:08:04 [WARNING] bbox_head.head.fc6.weight is not in pretrained model
2023-03-08 16:08:04 [WARNING] bbox_head.head.fc6.bias is not in pretrained model
2023-03-08 16:08:04 [WARNING] bbox_head.head.fc7.weight is not in pretrained model
2023-03-08 16:08:04 [WARNING] bbox_head.head.fc7.bias is not in pretrained model
2023-03-08 16:08:04 [WARNING] bbox_head.bbox_score.weight is not in pretrained model
2023-03-08 16:08:04 [WARNING] bbox_head.bbox_score.bias is not in pretrained model
2023-03-08 16:08:04 [WARNING] bbox_head.bbox_delta.weight is not in pretrained model
2023-03-08 16:08:04 [WARNING] bbox_head.bbox_delta.bias is not in pretrained model
2023-03-08 16:08:04 [INFO] There are 265/295 variables loaded into FasterRCNN.
2023-03-08 16:08:06 [INFO] [TRAIN] Epoch 1 finished, loss_rpn_cls=0.6888587, loss_rpn_reg=0.007873626, loss_bbox_cls=0.95418745, loss_bbox_reg=0.01720513, loss=1.6681249 .
2023-03-08 16:08:06 [INFO] Start to evaluate(total_samples=2, total_steps=2)...
---------------------------------------------------------------------------IndexError Traceback (most recent call last)/tmp/ipykernel_98/2582658238.py in
环境
-
请提供您使用的PaddlePaddle和PaddleX的版本号 PaddlePaddle 2.3.2 ；paddlex==2.1.0
-
请提供您使用的操作系统信息，如Linux/Windows/MacOS Windows11
-
请问您使用的Python版本是？ python 3.7
-
请问您使用的CUDA/cuDNN的版本号是？ V100 16GB
