def get_executor(use_gpu=True):
torch_module = MobileNetV2(n_class=27)
# print(torch_module)
if not os.path.exists(
"mobilenetv2_jester_online.pth.tar"): # checkpoint not downloaded
print('Downloading PyTorch checkpoint...')
import urllib.request
url = 'https://hanlab.mit.edu/projects/tsm/models/mobilenetv2_jester_online.pth.tar'
urllib.request.urlretrieve(url, './mobilenetv2_jester_online.pth.tar')
torch_module.load_state_dict(
torch.load("mobilenetv2_jester_online.pth.tar"))
torch_inputs = (torch.rand(1, 3, 224, 224), torch.zeros([1, 3, 56, 56]),
torch.zeros([1, 4, 28, 28]), torch.zeros([1, 4, 28, 28]),
torch.zeros([1, 8, 14, 14]), torch.zeros([1, 8, 14, 14]),
torch.zeros([1, 8, 14, 14]), torch.zeros([1, 12, 14, 14]),
torch.zeros([1, 12, 14, 14]), torch.zeros([1, 20, 7, 7]),
torch.zeros([1, 20, 7, 7]))
#summary(torch_module, torch_inputs)
if use_gpu:
target = 'cuda -model=tx2'
else:
target = 'llvm -device=arm_cpu -target=aarch64-linux-gnu' #-mcpu=cortex-a57
# target = tvm.target.arm_cpu('-target=aarch64-linux-gnu')
return torch2executor(torch_module, torch_inputs, target)
def get_executor():
torch_module = MobileNetV2(n_class=4)
torch_module.load_state_dict(torch.load("ckpt.best.pth.tar"), strict=False)
# torch_module.load_state_dict(torch.load("jester.pth.tar"))
torch_inputs = (torch.rand(1, 3, 224, 224),
torch.zeros([1, 3, 56, 56]),
torch.zeros([1, 4, 28, 28]),
torch.zeros([1, 4, 28, 28]),
torch.zeros([1, 8, 14, 14]),
torch.zeros([1, 8, 14, 14]),
torch.zeros([1, 8, 14, 14]),
torch.zeros([1, 12, 14, 14]),
torch.zeros([1, 12, 14, 14]),
torch.zeros([1, 20, 7, 7]),
torch.zeros([1, 20, 7, 7]))
return torch2executor(torch_module, torch_inputs, target='cuda')
def get_executor():
torch_module = MobileNetV2(n_class=27)
if not os.path.exists(
"mobilenetv2_jester_online.pth.tar"): # checkpoint not downloaded
print('Downloading PyTorch checkpoint...')
import urllib.request
url = 'https://hanlab.mit.edu/projects/tsm/models/mobilenetv2_jester_online.pth.tar'
urllib.request.urlretrieve(url, './mobilenetv2_jester_online.pth.tar')
torch_module.load_state_dict(
torch.load("mobilenetv2_jester_online.pth.tar"))
torch_inputs = (torch.rand(1, 3, 224, 224), torch.zeros([1, 3, 56, 56]),
torch.zeros([1, 4, 28, 28]), torch.zeros([1, 4, 28, 28]),
torch.zeros([1, 8, 14, 14]), torch.zeros([1, 8, 14, 14]),
torch.zeros([1, 8, 14, 14]), torch.zeros([1, 12, 14, 14]),
torch.zeros([1, 12, 14, 14]), torch.zeros([1, 20, 7, 7]),
torch.zeros([1, 20, 7, 7]))
return torch2executor(torch_module, torch_inputs, target='cuda')
def get_executor(use_gpu=True):
torch_module = MobileNetV2(n_class=27)
if not os.path.exists(
"mobilenetv2_jester_online.pth.tar"): # checkpoint not downloaded
print('Downloading PyTorch checkpoint...')
import urllib.request
url = 'https://file.lzhu.me/projects/tsm/models/mobilenetv2_jester_online.pth.tar'
urllib.request.urlretrieve(url, './mobilenetv2_jester_online.pth.tar')
torch_module.load_state_dict(
torch.load("mobilenetv2_jester_online.pth.tar"))
torch_inputs = (torch.rand(1, 3, 224, 224), torch.zeros([1, 3, 56, 56]),
torch.zeros([1, 4, 28, 28]), torch.zeros([1, 4, 28, 28]),
torch.zeros([1, 8, 14, 14]), torch.zeros([1, 8, 14, 14]),
torch.zeros([1, 8, 14, 14]), torch.zeros([1, 12, 14, 14]),
torch.zeros([1, 12, 14, 14]), torch.zeros([1, 20, 7, 7]),
torch.zeros([1, 20, 7, 7]))
if use_gpu:
target = 'cuda'
else:
target = 'llvm -mcpu=cortex-a72 -target=armv7l-linux-gnueabihf'
return torch2executor(torch_module, torch_inputs, target)
def get_executor():
model = MobileNetV2(n_class=27)
mobilenetv2_jester = torch.load('mobilenetv2_jester.pth.tar')['state_dict']
from collections import OrderedDict
new_state_dict = OrderedDict()
for k, v in mobilenetv2_jester.items():
name = k[7:]
if 'new_fc' in name:
name = name.replace('new_fc', 'classifier')
else:
if 'net' in name:
name = name.replace('net.', '')
name = name[11:]
new_state_dict[name] = v
model.load_state_dict(new_state_dict)
inputs = (torch.rand(1, 3, 224, 224), torch.zeros([1, 3, 56, 56]),
torch.zeros([1, 4, 28, 28]), torch.zeros([1, 4, 28, 28]),
torch.zeros([1, 8, 14, 14]), torch.zeros([1, 8, 14, 14]),
torch.zeros([1, 8, 14, 14]), torch.zeros([1, 12, 14, 14]),
torch.zeros([1, 12, 14,
14]), torch.zeros([1, 20, 7,
7]), torch.zeros([1, 20, 7, 7]))
return torch2executor(model, inputs, target='cuda')
import torch
dummy_input = torch.randn(1, 3, 224, 224)
state_dict = torch.load(r'./mobilenetv2_jester_online.pth.tar')
from mobilenet_v2_tsm import MobileNetV2
torch_module = MobileNetV2(n_class=27)
torch_module.load_state_dict(torch.load("mobilenetv2_jester_online.pth.tar"))
torch_module.eval()
torch_module.load_state_dict(state_dict)
shift_buffer = [torch.zeros([1, 3, 56, 56]),
torch.zeros([1, 4, 28, 28]),
torch.zeros([1, 4, 28, 28]),
torch.zeros([1, 8, 14, 14]),
torch.zeros([1, 8, 14, 14]),
torch.zeros([1, 8, 14, 14]),
torch.zeros([1, 12, 14, 14]),
torch.zeros([1, 12, 14, 14]),
torch.zeros([1, 20, 7, 7]),
torch.zeros([1, 20, 7, 7])]
# 因为需要偏移缓存区，所以有dummy_input, *shift_buffer两个输入，需要把它们组合成元组变成一个输入，
# 另外，还需加上opset_version=10，该方案来自：https://github.com/pytorch/fairseq/issues/1669#issuecomment-798972533
torch.onnx.export(torch_module, (dummy_input, *shift_buffer), "mobilenet_v2_tsm.onnx", opset_version=10, verbose=True)
def get_executor(use_gpu=True):
# torch_module = MobileNetV2(n_class=27)
# if not os.path.exists("mobilenetv2_jester_online.pth.tar"): # checkpoint not downloaded
# print('Downloading PyTorch checkpoint...')
# import urllib.request
# url = 'https://file.lzhu.me/projects/tsm/models/mobilenetv2_jester_online.pth.tar'
# urllib.request.urlretrieve(url, './mobilenetv2_jester_online.pth.tar')
torch_module = MobileNetV2(n_class=2)
# torch_module.load_state_dict(torch.load("ckpt.best.pth.tar"))
checkpoint = torch.load("ckpt.best.pth.tar")
checkpoint = checkpoint['state_dict']
# base_dict = {('base_model.' + k).replace('base_model.fc', 'new_fc'): v for k, v in list(checkpoint.items())}
base_dict = {
'.'.join(k.split('.')[1:]): v
for k, v in list(checkpoint.items())
}
replace_dict = {
'base_model.classifier.weight': 'new_fc.weight',
'base_model.classifier.bias': 'new_fc.bias',
}
# print(base_dict.keys())
for k in base_dict.keys():
k_new = k.replace('.net', '').replace('base_model.', '')
base_dict[k_new] = base_dict.pop(k)
# print(k, type(k))
# print('base_model.features.2.conv.4.running_var'.startswith('base_model.'))
# print('base_model.features.2.conv.4.running_var'.replace('base_model.', ''))
# # exit()
# if k.startswith('base_model.'):
# k_new = k.replace('base_model.', '')
# print(k_new)
# base_dict[k_new] = base_dict.pop(k)
# for k, v in replace_dict.items():
# if k in base_dict:
# base_dict[v] = base_dict.pop(k)
# print("\n\n\n")
# print(base_dict.keys())
for k in base_dict.keys():
k_new = k.replace('.net', '').replace('base_model.', '')
base_dict[k_new] = base_dict.pop(k)
# print("\n\n\n")
# print(base_dict.keys())
base_dict['classifier.weight'] = base_dict.pop('new_fc.weight')
base_dict['classifier.bias'] = base_dict.pop('new_fc.bias')
# exit()
# for k in base_dict.keys():
# print(k, type(k))
# print('base_model.features.2.conv.4.running_var'.startswith('base_model.'))
# print('base_model.features.2.conv.4.running_var'.replace('base_model.', ''))
# # exit()
# if k.startswith('base_model.'):
# k_new = k.replace('base_model.', '')
# print(k_new)
# base_dict[k_new] = base_dict.pop(k)
# print("\n\n\n")
# print(base_dict.keys())
torch_module.load_state_dict(base_dict)
# torch_module = torch.nn.DataParallel(torch_module.cuda())
# torch_module = torch_module.cuda()
# torch_module.eval()
# torch_module = TSN(2, 1, 'RGB',
# base_model='mobilenetv2',
# consensus_type='avg',
# img_feature_dim=256,
# pretrain='imagenet',
# # is_shift=False, shift_div=8, shift_place='blockres',
# is_shift=True, shift_div=8, shift_place='blockres',
# # non_local='_nl' in './checkpoint/TSM_HockeyFights_RGB_mobilenetv2_shift8_blockres_avg_segment8_e100/ckpt.best.pth.tar',
# non_local='_nl' in pt_path,
# )
# checkpoint = torch.load(
# # './checkpoint/TSM_HockeyFights_RGB_mobilenetv2_shift8_blockres_avg_segment8_e100/ckpt.best.pth.tar')
# pt_path)
# checkpoint = checkpoint['state_dict']
#
# # base_dict = {('base_model.' + k).replace('base_model.fc', 'new_fc'): v for k, v in list(checkpoint.items())}
# base_dict = {'.'.join(k.split('.')[1:]): v for k, v in list(checkpoint.items())}
# replace_dict = {'base_model.classifier.weight': 'new_fc.weight',
# 'base_model.classifier.bias': 'new_fc.bias',
# }
# for k, v in replace_dict.items():
# if k in base_dict:
# base_dict[v] = base_dict.pop(k)
# torch_module.load_state_dict(base_dict)
torch_inputs = (torch.rand(1, 3, 224, 224), torch.zeros([1, 3, 56, 56]),
torch.zeros([1, 4, 28, 28]), torch.zeros([1, 4, 28, 28]),
torch.zeros([1, 8, 14, 14]), torch.zeros([1, 8, 14, 14]),
torch.zeros([1, 8, 14, 14]), torch.zeros([1, 12, 14, 14]),
torch.zeros([1, 12, 14, 14]), torch.zeros([1, 20, 7, 7]),
torch.zeros([1, 20, 7, 7]))
# torch_inputs = (torch.rand(1, 3, 224, 224).cuda(),
# torch.zeros([1, 3, 56, 56]).cuda(),
# torch.zeros([1, 4, 28, 28]).cuda(),
# torch.zeros([1, 4, 28, 28]).cuda(),
# torch.zeros([1, 8, 14, 14]).cuda(),
# torch.zeros([1, 8, 14, 14]).cuda(),
# torch.zeros([1, 8, 14, 14]).cuda(),
# torch.zeros([1, 12, 14, 14]).cuda(),
# torch.zeros([1, 12, 14, 14]).cuda(),
# torch.zeros([1, 20, 7, 7]).cuda(),
# torch.zeros([1, 20, 7, 7]).cuda())
if use_gpu:
target = 'cuda'
else:
target = 'llvm -mcpu=cortex-a72 -target=armv7l-linux-gnueabihf'
# target = 'llvm -mcpu=cortex-a72 -target=armv7l-linux-gnueabihf'
return torch2executor(torch_module, torch_inputs, target)
