Created by: dmxj

faster-rcnn-r50模型训练，报错提示预训练模型地址不存在，使用的最新的master版本进行的训练。 基础训练参数：

architecture: FasterRCNN log_iter: 50 log_smooth_window: 50 max_iters: 20000 max_size: 1333 metric: COCO num_classes: 3 pretrain_weights: https://paddle-imagenet-models-name.bj.bcebos.com/ResNet50_vd_pretrained.tar save_dir: output snapshot_iter: 400 target_size: 800 use_gpu: true weights: '' FasterRCNN: backbone: ResNet bbox_assigner: BBoxAssigner bbox_head: BBoxHead fpn: FPN roi_extractor: FPNRoIAlign rpn_head: FPNRPNHead ResNet: dcn_v2_stages:[3,4,5] depth: 50 gcb_stages: [] norm_type: affine_channel variant: b

报错信息：

max_size: 1333 metric: COCO num_classes: 3 pretrain_weights: https://paddle-imagenet-models-name.bj.bcebos.com/ResNet50_vd_pretrained.tar save_dir: output snapshot_iter: 400 target_size: 800 use_gpu: true loading annotations into memory... Done (t=0.00s) creating index... index created! 2020-02-22 01:39:44,809-INFO: 71 samples in file /root/datasets/jingyan/dibu/coco/annotations/val.json 2020-02-22 01:39:44,809-INFO: places would be ommited when DataLoader is not iterable W0222 01:39:46.168161 46917 device_context.cc:236] Please NOTE: device: 0, CUDA Capability: 70, Driver API Version: 9.2, Runtime API Version: 9.0 W0222 01:39:46.171754 46917 device_context.cc:244] device: 0, cuDNN Version: 7.4. 2020-02-22 01:39:47,616-INFO: Load model and fuse batch norm if have from https://paddle-imagenet-models-name.bj.bcebos.com/ResNet50_vd_pretrained.tar... 2020-02-22 01:39:47,616-INFO: Found /root/.cache/paddle/weights/ResNet50_vd_pretrained 2020-02-22 01:39:47,621-INFO: Loading parameters from /root/.cache/paddle/weights/ResNet50_vd_pretrained... Traceback (most recent call last): File "/root/opensource/PaddleDetection/tools/train.py", line 323, in main() File "/root/opensource/PaddleDetection/tools/train.py", line 191, in main checkpoint.load_and_fusebn(exe, train_prog, cfg.pretrain_weights) File "/root/opensource/PaddleDetection/ppdet/utils/checkpoint.py", line 275, in load_and_fusebn load_params(exe, prog, path) File "/root/opensource/PaddleDetection/ppdet/utils/checkpoint.py", line 125, in load_params fluid.load(prog, path, executor=exe) File "/opt/conda/lib/python3.6/site-packages/paddle/fluid/io.py", line 1577, in load "Parameter file [{}] not exits".format(parameter_file_name) AssertionError: Parameter file [/root/.cache/paddle/weights/ResNet50_vd_pretrained.pdparams] not exits

paddlepaddle-gpu版本为1.6.2.post97，操作系统为ubuntu，前几天还是可以正常训练的，今天就报了这个错