Unexpected Key error when trying to load the COS.ckpt checkpoint for testing it on miniImageNet test data.
Deepayan137 opened this issue · comments
Hello.
I am trying to test the COS.ckpt model on miniImageNet test data, the link to which is provided in the readme file. I have made the required changes in the set_config_COSOC.py as mentioned and I am calling the bash run.sh. I am pasting the error that I am seeing. Any help would be much appreciated.
RuntimeError: Error(s) in loading state_dict for COSOC:
Missing key(s) in state_dict: "backbone.layer1.conv1.weight", "backbone.layer1.bn1.weight", "backbone.layer1.bn1.bias", "backbone.layer1.bn1.running_mean", "backbone.layer1.bn1.running_var", "backbone.layer1.conv2.weight", "backbone.layer1.bn2.weight", "backbone.layer1.bn2.bias", "backbone.layer1.bn2.running_mean", "backbone.layer1.bn2.running_var", "backbone.layer1.conv3.weight", "backbone.layer1.bn3.weight", "backbone.layer1.bn3.bias", "backbone.layer1.bn3.running_mean", "backbone.layer1.bn3.running_var", "backbone.layer1.downsample.0.weight", "backbone.layer1.downsample.1.weight", "backbone.layer1.downsample.1.bias", "backbone.layer1.downsample.1.running_mean", "backbone.layer1.downsample.1.running_var", "backbone.layer2.conv1.weight", "backbone.layer2.bn1.weight", "backbone.layer2.bn1.bias", "backbone.layer2.bn1.running_mean", "backbone.layer2.bn1.running_var", "backbone.layer2.conv2.weight", "backbone.layer2.bn2.weight", "backbone.layer2.bn2.bias", "backbone.layer2.bn2.running_mean", "backbone.layer2.bn2.running_var", "backbone.layer2.conv3.weight", "backbone.layer2.bn3.weight", "backbone.layer2.bn3.bias", "backbone.layer2.bn3.running_mean", "backbone.layer2.bn3.running_var", "backbone.layer2.downsample.0.weight", "backbone.layer2.downsample.1.weight", "backbone.layer2.downsample.1.bias", "backbone.layer2.downsample.1.running_mean", "backbone.layer2.downsample.1.running_var", "backbone.layer3.conv1.weight", "backbone.layer3.bn1.weight", "backbone.layer3.bn1.bias", "backbone.layer3.bn1.running_mean", "backbone.layer3.bn1.running_var", "backbone.layer3.conv2.weight", "backbone.layer3.bn2.weight", "backbone.layer3.bn2.bias", "backbone.layer3.bn2.running_mean", "backbone.layer3.bn2.running_var", "backbone.layer3.conv3.weight", "backbone.layer3.bn3.weight", "backbone.layer3.bn3.bias", "backbone.layer3.bn3.running_mean", "backbone.layer3.bn3.running_var", "backbone.layer3.downsample.0.weight", "backbone.layer3.downsample.1.weight", "backbone.layer3.downsample.1.bias", "backbone.layer3.downsample.1.running_mean", "backbone.layer3.downsample.1.running_var", "backbone.layer4.conv1.weight", "backbone.layer4.bn1.weight", "backbone.layer4.bn1.bias", "backbone.layer4.bn1.running_mean", "backbone.layer4.bn1.running_var", "backbone.layer4.conv2.weight", "backbone.layer4.bn2.weight", "backbone.layer4.bn2.bias", "backbone.layer4.bn2.running_mean", "backbone.layer4.bn2.running_var", "backbone.layer4.conv3.weight", "backbone.layer4.bn3.weight", "backbone.layer4.bn3.bias", "backbone.layer4.bn3.running_mean", "backbone.layer4.bn3.running_var", "backbone.layer4.downsample.0.weight", "backbone.layer4.downsample.1.weight", "backbone.layer4.downsample.1.bias", "backbone.layer4.downsample.1.running_mean", "backbone.layer4.downsample.1.running_var", "classifier.scale_cls", "classifier.L.weight_g", "classifier.L.weight_v".
Unexpected key(s) in state_dict: "layer1.conv1.weight", "layer1.bn1.weight", "layer1.bn1.bias", "layer1.bn1.running_mean", "layer1.bn1.running_var", "layer1.bn1.num_batches_tracked", "layer1.conv2.weight", "layer1.bn2.weight", "layer1.bn2.bias", "layer1.bn2.running_mean", "layer1.bn2.running_var", "layer1.bn2.num_batches_tracked", "layer1.conv3.weight", "layer1.bn3.weight", "layer1.bn3.bias", "layer1.bn3.running_mean", "layer1.bn3.running_var", "layer1.bn3.num_batches_tracked", "layer1.downsample.0.weight", "layer1.downsample.1.weight", "layer1.downsample.1.bias", "layer1.downsample.1.running_mean", "layer1.downsample.1.running_var", "layer1.downsample.1.num_batches_tracked", "layer2.conv1.weight", "layer2.bn1.weight", "layer2.bn1.bias", "layer2.bn1.running_mean", "layer2.bn1.running_var", "layer2.bn1.num_batches_tracked", "layer2.conv2.weight", "layer2.bn2.weight", "layer2.bn2.bias", "layer2.bn2.running_mean", "layer2.bn2.running_var", "layer2.bn2.num_batches_tracked", "layer2.conv3.weight", "layer2.bn3.weight", "layer2.bn3.bias", "layer2.bn3.running_mean", "layer2.bn3.running_var", "layer2.bn3.num_batches_tracked", "layer2.downsample.0.weight", "layer2.downsample.1.weight", "layer2.downsample.1.bias", "layer2.downsample.1.running_mean", "layer2.downsample.1.running_var", "layer2.downsample.1.num_batches_tracked", "layer3.conv1.weight", "layer3.bn1.weight", "layer3.bn1.bias", "layer3.bn1.running_mean", "layer3.bn1.running_var", "layer3.bn1.num_batches_tracked", "layer3.conv2.weight", "layer3.bn2.weight", "layer3.bn2.bias", "layer3.bn2.running_mean", "layer3.bn2.running_var", "layer3.bn2.num_batches_tracked", "layer3.conv3.weight", "layer3.bn3.weight", "layer3.bn3.bias", "layer3.bn3.running_mean", "layer3.bn3.running_var", "layer3.bn3.num_batches_tracked", "layer3.downsample.0.weight", "layer3.downsample.1.weight", "layer3.downsample.1.bias", "layer3.downsample.1.running_mean", "layer3.downsample.1.running_var", "layer3.downsample.1.num_batches_tracked", "layer4.conv1.weight", "layer4.bn1.weight", "layer4.bn1.bias", "layer4.bn1.running_mean", "layer4.bn1.running_var", "layer4.bn1.num_batches_tracked", "layer4.conv2.weight", "layer4.bn2.weight", "layer4.bn2.bias", "layer4.bn2.running_mean", "layer4.bn2.running_var", "layer4.bn2.num_batches_tracked", "layer4.conv3.weight", "layer4.bn3.weight", "layer4.bn3.bias", "layer4.bn3.running_mean", "layer4.bn3.running_var", "layer4.bn3.num_batches_tracked", "layer4.downsample.0.weight", "layer4.downsample.1.weight", "layer4.downsample.1.bias", "layer4.downsample.1.running_mean", "layer4.downsample.1.running_var", "layer4.downsample.1.num_batches_tracked".
Also, pasting the set_config_COSOC.py file
from sacred import Experiment
import yaml
import time
import os
ex = Experiment("COSOC", save_git_info=False)
@ex.config
def config():
config_dict = {}
# whether load pretrained model. This is is different from resume_from_checkpoint that loads everything from a breakpoint.
config_dict["load_pretrained"] = True
#if training, set to False
config_dict["is_test"] = True
if config_dict["is_test"]:
#if testing, specify the total rounds of testing. Default: 5
config_dict["num_test"] = 5
config_dict["load_pretrained"] = True
#specify pretrained path for testing.
if config_dict["load_pretrained"]:
config_dict["pre_trained_path"] = "results/COS.ckpt"
# config_dict["pre_trained_path"] = "../results/Exemplar/1000epoch/version_0/checkpoints/epoch=864-step=129749.ckpt"
# config_dict["pre_trained_path"] = "../results/ProtoNet/test1gpu/version_0/checkpoints/epoch=59-step=29999.ckpt"
#only load the backbone.
config_dict["load_backbone_only"] = False
#Specify the model name, which should match the name of file
#that contains the LightningModule
config_dict["model_name"] = "COSOC"
#whether to use multiple GPUs
multi_gpu = False
if config_dict["is_test"]:
multi_gpu = False
#The seed
seed = 10
config_dict["seed"] = seed
#The logging dirname: logdir/exp_name/
log_dir = "results/"
exp_name = "COSOC/test"
#Three components of a Lightning Running System
trainer = {}
data = {}
model = {}
################trainer configuration###########################
###important###
#debugging mode
trainer["fast_dev_run"] = False
if multi_gpu:
trainer["accelerator"] = "ddp"
trainer["sync_batchnorm"] = True
trainer["gpus"] = [2,3]
trainer["plugins"] = [{"class_path": "plugins.modified_DDPPlugin"}]
else:
trainer["accelerator"] = None
trainer["gpus"] = [2]
trainer["sync_batchnorm"] = False
# whether resume from a given checkpoint file
trainer["resume_from_checkpoint"] = None # example: "../results/ProtoNet/version_11/checkpoints/epoch=2-step=1499.ckpt"
# The maximum epochs to run
trainer["max_epochs"] = 20
# potential functionalities added to the trainer.
trainer["callbacks"] = [{"class_path": "pytorch_lightning.callbacks.LearningRateMonitor",
"init_args": {"logging_interval": "step"}
},
{"class_path": "pytorch_lightning.callbacks.ModelCheckpoint",
"init_args":{"verbose": True, "save_last": True, "monitor": "val/acc", "mode": "max"}
},
{"class_path": "callbacks.SetSeedCallback",
"init_args":{"seed": seed, "is_DDP": multi_gpu}
}]
###less important###
num_gpus = trainer["gpus"] if isinstance(trainer["gpus"], int) else len(trainer["gpus"])
trainer["logger"] = {"class_path":"pytorch_lightning.loggers.TensorBoardLogger",
"init_args": {"save_dir": log_dir,"name": exp_name}
}
trainer["replace_sampler_ddp"] = False
##################shared model and datamodule configuration###########################
#important
test_shot = 1
num_patch = 7
#less important
per_gpu_val_batchsize = 1
per_gpu_test_batchsize = 1
val_way = 5
test_way = 5
val_shot = 1
num_query = 15
##################datamodule configuration###########################
#important
#The name of dataset, which should match the name of file
#that contains the datamodule.
data["train_dataset_name"] = "miniImageNet_prob_crop"
data["val_test_dataset_name"] = "miniImageNet_multi_crop"
data["train_dataset_params"] = {"feature_image_and_crop_id":"../result/feature_image_and_crop_id.pkl","position_list":"../result/position_list.npy"}
data["train_data_root"] = "../BF3S-master/data/mini_imagenet_split/images"
data["val_test_data_root"] = "../BF3S-master/data/mini_imagenet_split/images"
data["val_test_dataset_params"] = {"num_patch":num_patch}
#determine whether meta-learning.
data["train_batchsize"] = 128
data["train_num_workers"] = 8
#the number of tasks
data["val_num_task"] = 600
data["test_num_task"] = 2000
#less important
data["num_gpus"] = num_gpus
data["val_batchsize"] = num_gpus*per_gpu_val_batchsize
data["test_batchsize"] = num_gpus*per_gpu_test_batchsize
data["test_shot"] = test_shot
data["val_num_workers"] = 8
data["is_DDP"] = True if multi_gpu else False
data["val_way"] = val_way
data["test_way"] = test_way
data["val_shot"] = val_shot
data["num_query"] = num_query
data["drop_last"] = False
data["is_meta"] = False
##################model configuration###########################
#important
#The name of feature extractor, which should match the name of file
#that contains the model.
model["backbone_name"] = "resnet12"
#the initial learning rate
model["lr"] = 0.005*data["train_batchsize"]/128
#less important
model["SOC_params"] = {"num_patch": num_patch, "alpha": 1.0, "beta": 0.8}
model["val_way"] = val_way
model["test_way"] = test_way
model["val_shot"] = val_shot
model["test_shot"] = test_shot
model["num_query"] = num_query
model["val_batch_size_per_gpu"] = per_gpu_val_batchsize
model["test_batch_size_per_gpu"] = per_gpu_test_batchsize
model["weight_decay"] = 5e-4
#The name of optimization scheduler
model["decay_scheduler"] = "cosine"
model["optim_type"] = "sgd"
model["num_classes"] = 64
config_dict["trainer"] = trainer
config_dict["data"] = data
config_dict["model"] = model
@ex.automain
def main(_config):
config_dict = _config["config_dict"]
file_ = 'config/config.yaml'
stream = open(file_, 'w')
yaml.safe_dump(config_dict, stream=stream,default_flow_style=False)
Hi! you may notice that the expected keys and missing keys are quite similar, with the difference of the prefix "backbone". The loading code is in lines 96-102 of run.py, you can turn config_dict["load_backbone_only"] in the config file to True and try to modify the keys in the "state" variable to match the two keys.
Thank you for the quick response. Yes, turning the load_backbone_only flag to True and commenting out line 101 did the trick. I am closing this issue.