def test_save_load():
np.random.seed(123)
a_np = np.random.randn(3, 2).astype(np.float32)
a = PyTorch.asFloatTensor(a_np)
print('a', a)
filename = '/tmp/foo.t7' # TODO: should use tempfile to get this
PyTorchAug.save(filename, a)
b = PyTorchAug.load(filename)
print('type(b)', type(b))
print('b', b)
assert np.abs(a_np - b.asNumpyTensor()).max() < 1e-4
def load(self, path):
# dont remove this line. This is important to init the PyTorch nn
self.init_modules()
print('loading from: ' + path)
model = PyTorchAug.load(path)
print(model)
return model
def __init__(self, *args):
_fromLua = False
if len(args) >= 1:
if args[0] == '__FROMLUA__':
_fromLua = True
args = args[1:]
# print('LuaWrapper.__init__', lua_classname, 'fromLua', _fromLua, 'args', args)
self.luaclass = lua_classname
if not _fromLua:
PyTorchAug.LuaClass.__init__(self, splitName, *args)
else:
self.__dict__['__objectId'] = PyTorchAug.getNextObjectId()
def __init__(self, *args):
_fromLua = False
if len(args) >= 1:
if args[0] == '__FROMLUA__':
_fromLua = True
args = args[1:]
# print('LuaWrapper.__init__', lua_classname, 'fromLua', _fromLua, 'args', args)
self.luaclass = lua_classname
if not _fromLua:
PyTorchAug.LuaClass.__init__(self, splitName, *args)
else:
self.__dict__['__objectId'] = PyTorchAug.getNextObjectId()
lua.call(1, 0)
lua = PyTorchAug.lua
lua.getGlobal("require")
lua.pushString('modules/SelectiveGradientFilter')
lua.call(1, 0)
dim_hidden = 200
feature_maps = 96
filter_size = 5
colorchaPyTorchAugels = 1
# ipdb.set_trace()
encoder = PyTorchAug.Sequential()
encoder.add(
PyTorchAug.SpatialConvolution(colorchaPyTorchAugels, feature_maps,
filter_size, filter_size))
encoder.add(PyTorchAug.SpatialMaxPooling(2, 2, 2, 2))
encoder.add(PyTorchAug.Threshold(0, 1e-6))
encoder.add(
PyTorchAug.SpatialConvolution(feature_maps, feature_maps / 2, filter_size,
filter_size))
encoder.add(PyTorchAug.SpatialMaxPooling(2, 2, 2, 2))
encoder.add(PyTorchAug.Threshold(0, 1e-6))
encoder.add(
net.params['rpn_cls_score'][0].data.reshape(18, 512))
transferB(tnet["rpn_cls_score"].bias, net.params['rpn_cls_score'][1].data)
tnet["rpn_bbox_pred"] = nn.SpatialConvolutionMM(512, 36, 1, 1, 1, 1, 0, 0)
transferW(tnet["rpn_bbox_pred"].weight,
net.params['rpn_bbox_pred'][0].data.reshape(36, 512))
transferB(tnet["rpn_bbox_pred"].bias, net.params['rpn_bbox_pred'][1].data)
tnet["fc6"] = nn.Linear(25088, 4096)
transferW(tnet["fc6"].weight, net.params['fc6'][0].data)
transferB(tnet["fc6"].bias, net.params['fc6'][1].data)
tnet["relu6"] = nn.ReLU()
tnet["drop6"] = nn.Dropout(0.5)
tnet["fc7"] = nn.Linear(4096, 4096)
transferW(tnet["fc7"].weight, net.params['fc7'][0].data)
transferB(tnet["fc7"].bias, net.params['fc7'][1].data)
tnet["relu7"] = nn.ReLU()
tnet["drop7"] = nn.Dropout(0.5)
tnet["cls_score"] = nn.Linear(4096, 21)
transferW(tnet["cls_score"].weight, net.params['cls_score'][0].data)
transferB(tnet["cls_score"].bias, net.params['cls_score'][1].data)
tnet["bbox_pred"] = nn.Linear(4096, 84)
transferW(tnet["bbox_pred"].weight, net.params['bbox_pred'][0].data)
transferB(tnet["bbox_pred"].bias, net.params['bbox_pred'][1].data)
print net.params
PyTorchAug.save("faster_rcnn_2.t7", tnet)
# print('res', res)
return res
def cuda_from_double(self):
# print('cl')
res = PyCudaTorch.DoubleTensorToCudaTensor(self)
# print('res', res)
return res
FloatTensor.cuda = cuda_from_float
DoubleTensor.cuda = cuda_from_double
#PyTorchAug.
#def Linear_cl(self):
# print('Linear_cl')
# print('self', self)
# self.cl()
# return self
## import PyTorch
#Linear.cl = Linear_cl
print('dir(PyTorchAug)', dir(PyTorchAug))
print('dir(PyCudaTorch)', dir(PyCudaTorch))
PyTorchAug.cythonClasses['torch.CudaTensor'] = {'popFunction': PyCudaTorch.cyPopCudaTensor}
PyTorchAug.populateLuaClassesReverse()
PyTorchAug.pushFunctionByPythonClass[PyCudaTorch.CudaTensor] = PyCudaTorch.cyPushCudaTensor
def cl(self):
# print('cl')
res = PyClTorch.FloatTensorToClTensor(self)
# print('res', res)
return res
FloatTensor.cl = cl
#PyTorchAug.
#def Linear_cl(self):
# print('Linear_cl')
# print('self', self)
# self.cl()
# return self
## import PyTorch
#Linear.cl = Linear_cl
print('dir(PyTorchAug)', dir(PyTorchAug))
print('dir(PyClTorch)', dir(PyClTorch))
PyTorchAug.cythonClasses['torch.ClTensor'] = {
'popFunction': PyClTorch.cyPopClTensor
}
PyTorchAug.populateLuaClassesReverse()
PyTorchAug.pushFunctionByPythonClass[
PyClTorch.ClTensor] = PyClTorch.cyPushClTensor
transferW(tnet["conv5_3"].weight,
net.params['conv5_3'][0].data.reshape(512, 4608))
transferB(tnet["conv5_3"].bias, net.params['conv5_3'][1].data)
tnet["relu5_3"] = nn.ReLU()
tnet["rpn_conv/3x3"] = nn.SpatialConvolutionMM(512, 512, 3, 3, 1, 1, 1, 1)
tnet["rpn_relu/3x3"] = nn.ReLU()
tnet["rpn_cls_score"] = nn.SpatialConvolutionMM(512, 18, 1, 1, 1, 1, 0, 0)
tnet["rpn_bbox_pred"] = nn.SpatialConvolutionMM(512, 36, 1, 1, 1, 1, 0, 0)
tnet["fc6"] = nn.Linear(25088, 4096)
transferW(tnet["fc6"].weight, net.params['fc6'][0].data)
transferB(tnet["fc6"].bias, net.params['fc6'][1].data)
tnet["relu6"] = nn.ReLU()
tnet["drop6"] = nn.Dropout(0.5)
tnet["fc7"] = nn.Linear(4096, 4096)
transferW(tnet["fc7"].weight, net.params['fc7'][0].data)
transferB(tnet["fc7"].bias, net.params['fc7'][1].data)
tnet["relu7"] = nn.ReLU()
tnet["drop7"] = nn.Dropout(0.5)
tnet["cls_score"] = nn.Linear(4096, 21)
tnet["bbox_pred"] = nn.Linear(4096, 84)
print net.params
PyTorchAug.save("vgg16.t7", tnet)
transferB(tnet["rpn_bbox_pred"].bias, net.params['rpn_bbox_pred'][1].data)
tnet["fc6"] = nn.Linear(18432,4096) #512*6*6
print "fc6"
transferW(tnet["fc6"].weight, net.params['fc6'][0].data)
transferB(tnet["fc6"].bias, net.params['fc6'][1].data)
tnet["relu6"] = nn.ReLU()
tnet["drop6"] = nn.Dropout(0.5)
tnet["fc7"] = nn.Linear(4096,1024)
print "fc7"
transferW(tnet["fc7"].weight, net.params['fc7'][0].data)
transferB(tnet["fc7"].bias, net.params['fc7'][1].data)
tnet["relu7"] = nn.ReLU()
tnet["drop7"] = nn.Dropout(0.5)
tnet["cls_score"] = nn.Linear(1024,21)
if args.rpn:
print "cls_score"
transferW(tnet["cls_score"].weight, net.params['cls_score'][0].data)
transferB(tnet["cls_score"].bias, net.params['cls_score'][1].data)
tnet["bbox_pred"] = nn.Linear(1024,84)
if args.rpn:
print "bbox_pred"
transferW(tnet["bbox_pred"].weight, net.params['bbox_pred'][0].data)
transferB(tnet["bbox_pred"].bias, net.params['bbox_pred'][1].data)
print net.params
PyTorchAug.save(args.out,tnet)
def save(self, model, path):
PyTorchAug.save(path, model)
