drive.google

[mnlvc]

Hello! No access to google drive with the model pretrained weights

[bk202]

Hi @PushpakBhoge , I'm also running into permission issues when running download_model_weights.py, can you please share the permission for downloading the model? Thanks in advance

[PushpakBhoge]

HI @bk202 @mnlvc I checked the drive seems like the files got deleted during recent clean-up policy of google
unfortunately I did not find any backup of the weight since it was long back

you will have to download weights from the original repo extract those layer by layer and then load those in the new network
refer this snippet on how to load in model.

Face_Recognition_TF/Inference/FaceRecognizer.py

Lines 36 to 61 in 0e6ad2d

	def load_model(self):
	"""
	Arguments:
	None
	Output:
	a model instance
	This method is not meant to be called outside class
	"""
	# load model from source
	model = InceptionResNetV1()
	# Load weights layer by layer
	layer_files = os.listdir(self.WEIGHT_BASE)
	for i, layer in enumerate(model.layers):
	weight_files = [x for x in layer_files if x.split(".")[0]==layer.name]
	for weight_file in weight_files:
	files_loaded = np.load(os.path.join(self.WEIGHT_BASE, weight_file))
	weights_for_layer = []
	for file in files_loaded:
	weights_for_layer.append(files_loaded[file])
	try:
	layer.set_weights(weights_for_layer)
	except:
	pass
	return model

when you export weights layer by layer save the .npy file name them as layer names like {layer_name}.npy refer this for exporting https://stackoverflow.com/questions/43715047/how-do-i-get-the-weights-of-a-layer-in-keras

I am planning to rewrite this repository completely in 1-2 months with better code architecture, features and custom training scripts etc. so stay tuned

P.S. the current code might not work on the latest version of tensorflow

[Mohit-robo]

In that case could you please share the training code

[jankaWIS]

HI @bk202 @mnlvc I checked the drive seems like the files got deleted during recent clean-up policy of google unfortunately I did not find any backup of the weight since it was long back

you will have to download weights from the original repo extract those layer by layer and then load those in the new network refer this snippet on how to load in model.

Face_Recognition_TF/Inference/FaceRecognizer.py

Lines 36 to 61 in 0e6ad2d
def load_model(self):
"""
Arguments:
None
Output:
a model instance
This method is not meant to be called outside class
"""
# load model from source
model = InceptionResNetV1()

     # Load weights layer by layer 
     layer_files = os.listdir(self.WEIGHT_BASE) 
     for i, layer in enumerate(model.layers): 
         weight_files = [x for x in layer_files if x.split(".")[0]==layer.name] 
         for weight_file in weight_files: 
             files_loaded = np.load(os.path.join(self.WEIGHT_BASE, weight_file)) 
             weights_for_layer = [] 
             for file in files_loaded: 
                 weights_for_layer.append(files_loaded[file]) 
         try: 
             layer.set_weights(weights_for_layer) 
         except: 
             pass 

     return model 

when you export weights layer by layer save the .npy file name them as layer names like {layer_name}.npy refer this for exporting https://stackoverflow.com/questions/43715047/how-do-i-get-the-weights-of-a-layer-in-keras

I am planning to rewrite this repository completely in 1-2 months with better code architecture, features and custom training scripts etc. so stay tuned

P.S. the current code might not work on the latest version of tensorflow

Hi,

Did you manage to update the repo or maybe create a new one with updated content?

Thanks.

[PushpakBhoge]

Hi @jankaWIS
I sincerely apologize, but due to commitments with my employer and other ongoing responsibilities, I haven’t been able to find the time to continue pursuing this personal project.

[PushpakBhoge]

Hi @Mohit-robo,
Apologies for the significantly delayed response. At the time, I did not develop any training code for the project. It was solely an application built on top of a pre-trained model. The original repository is referenced in the README.md file.