@wang21jun Hello, I have a small question, In this project you used cosine similarity for caculate score between 2 feature. This score can be negative so I wanna normalize it between 0 and 1. Can you share the solution to do it ? Thanks

Hello,

Let's explain the issue with below samples:

I use add_mask_one script, is_aug is False but sometimes, it somehow changes the whole image content. I just want to wear a mask without effect to my image quality.

Hi. @wang21jun . Thank for your contribute. I have a question about the model using for recognition, I saw many pretrain model in https://github.com/JDAI-CV/FaceX-Zoo/tree/main/training_mode but I dont know what model you used ? . In config file I saw this model name is face_recognition_1.0 but I dont know the architecture of this model ( backbone, head v..v). Can you share me the details of this model ? Thanks.

This is a great repo, thanks for sharing.

I have some problems here. I want to make some new mask UV texture map templates for myself. Could you please provide the uv_mask.png and process scripts? thank you.

识别模型：ReXNetV1 转换脚本： model_path="out_dir/Epoch_249.pt" onnx_path="models/ReXNetV1.onnx"

class OnnxConvertor: def init(self): pass def convert(self, model, output_path): img = np.random.randint(0, 255, size=(112,112,3), dtype=np.uint8) img = img[:,:,::-1].astype(np.float32) img = (img.transpose((2, 0, 1)) - 127.5) * 0.0078125 img = torch.from_numpy(img).unsqueeze(0).float() torch.onnx.export( model, img, output_path, keep_initializers_as_inputs=False, verbose=False, opset_version=12 ) # batch inference. onnx_model = onnx.load(output_path) graph = onnx_model.graph graph.input[0].type.tensor_type.shape.dim[0].dim_param = 'None' onnx.save(onnx_model, output_path)

if name == 'main': model = ReXNetV1(input_ch=16, final_ch=180, width_mult=1.0, depth_mult=1.0, use_se=0, se_ratio=12, out_h=7, out_w=7, feat_dim=512, dropout_ratio=0.2) #model.load_state_dict(torch.load(model_path)["state_dict"]) model_dict = model.state_dict() pretrained_dict = torch.load(model_path)['state_dict'] new_pretrained_dict = {} for k in model_dict: if k.startswith("features."): print(k) new_pretrained_dict[k] = pretrained_dict[k] model_dict.update(new_pretrained_dict) model.load_state_dict(model_dict) model.eval() #dummy_input = torch.randn(1, 3, 112, 112) #torch.onnx.export(model, dummy_input, onnx_path, opset_version=11) convertor = OnnxConvertor() convertor.convert(model, onnx_path)

使用3.1 Experiments of SOTA backbones下的模型时，发现有些模型可以加载成功有些会报错 我的pytorch为1.1.0 GPU版本

按照如下方式加载MobileFaceNet模型成功 model = MobileFaceNet(512, 7, 7)
model = nn.DataParallel(model) model.load_state_dict(torch.load(‘model.pt', False)

但是按照同样方式加载Resnet50-ir模型报错“invalid load key, '\xff'.” model = Resnet(152, 0.4) model = nn.DataParallel(model) model.load_state_dict(torch.load(‘model.pt’, False)

请问一下要怎么样才能正确加载Resnet50-ir模型？

i downloaded the pretrained model Attention-92(MX) and the LFW acc announced as 99.82% but as i used the test_lfw.py to get the result , only 85.9% accuracy achieved.

the log is below:

backbone param:
{'stage1_modules': 1, 'stage2_modules': 2, 'stage3_modules': 3, 'feat_dim': 512, 'out_h': 7, 'out_w': 7}
100%|██████████| 828/828 [06:51<00:00,  2.26it/s]
6000it [00:00, 22128.20it/s]
+-------------+---------------+----------------------+
|  model_name | mean accuracy |    standard error    |
+-------------+---------------+----------------------+
| Epoch_17.pt |     0.859     | 0.007375300789011594 |
+-------------+---------------+----------------------+

Process finished with exit code 0

any idea about it ?

Hi I tried with "Semi-Siamese Training" and found a config named queue_size in the training_mode/head_config.yaml.

SST_Prototype: queue_size: 16384 #is divisible by batch size.

When I read in the paper "Semi-Siamese Training for Shallow Face Learning", author mentioned that

The size of the gallery queue depends on the number of classes in training datasets, so we empirically set it as 16,384 for shallow and deep data, and 2,560 for QMULSurvFace.

May I ask how to choose appropriate queue_size with given a custom dataset

Thank you.

Hi,

I found the trained models in https://github.com/JDAI-CV/FaceX-Zoo/tree/main/training_mode, where you offer Attention-56 and Attention-92(MX).

However, when I looked at backbone codes backbone and backboneconf, I only saw a single model and configuration.

As I am gonna use your models as feature extraction backbones, I am wondering how to initialize those two backbones (Attention-56 and Attention-92(MX)) properly.

Looking forward to your reply. Thanks!

I start the project on cpu, but getting an error module must have its parameters and buffers on device cuda:0 (device_ids[0]) but found one of them on device: cpu Please tell me how to fix it? I changed loaders with 'cuda:0' -> 'cpu', also i added map_location parameter, but it still doesn't help

add render.pyx and setup.py under FaceX-Zoo/addition_module/face_mask_adding/FMA-3D/utils/cython to implement render

before python add_mask_one.py 11.24s user 2.37s system 117% cpu 11.593 total now python add_mask_one.py 3.25s user 2.34s system 159% cpu 3.506 total

An error happen while I try to train the ResNet model with my custom dataset on GoogleColab /content/FaceX-Zoo/training_mode/semi-siamese_training/train_vg.sh: 3: /content/FaceX-Zoo/training_mode/semi-siamese_training/train_vg.sh: mkdire: not found INFO 2022-10-20 04:51:36 train.py: 189] Start optimization. INFO 2022-10-20 04:51:36 train.py: 190] Namespace(alpha=0.999, backbone_conf_file='/content/FaceX-Zoo/training_mode/backbone_conf.yaml', backbone_type='ResNet', batch_size=8, data_root='/content/VN_celeb', epoches=250, evaluate=False, head_conf_file='/content/FaceX-Zoo/training_mode/head_conf.yaml', head_type='ArcFace', log_dir='log', lr=0.1, milestones=[60, 100, 120], momentum=0.9, out_dir='log/model', pretrain_model='/content/resnet152.pt', print_freq=20, resume=False, save_freq=10, step='60,100,120', tensorboardx_logdir='log/board', test_data_conf_file=None, test_set='LFW', train_file='/content/FaceX-Zoo/training_mode/train_list.txt', writer=<tensorboardX.writer.SummaryWriter object at 0x7fe18ac7ba50>) backbone param: {'depth': 152, 'drop_ratio': 0.4, 'net_mode': 'ir_se', 'feat_dim': 512, 'out_h': 7, 'out_w': 7} head param: {'feat_dim': 512, 'num_class': 72778, 'margin_arc': 0.35, 'margin_am': 0.0, 'scale': 32} Valid ids: 1020. /usr/local/lib/python3.7/dist-packages/torch/utils/data/dataloader.py:566: UserWarning: This DataLoader will create 4 worker processes in total. Our suggested max number of worker in current system is 2, which is smaller than what this DataLoader is going to create. Please be aware that excessive worker creation might get DataLoader running slow or even freeze, lower the worker number to avoid potential slowness/freeze if necessary. cpuset_checked)) Traceback (most recent call last): File "/content/FaceX-Zoo/training_mode/semi-siamese_training/train.py", line 191, in train(args) File "/content/FaceX-Zoo/training_mode/semi-siamese_training/train.py", line 125, in train prototype, optimizer, criterion, epoch, conf, loss_meter) File "/content/FaceX-Zoo/training_mode/semi-siamese_training/train.py", line 77, in train_one_epoch images1_probe, images2_gallery, images2_probe, images1_gallery, id_indexes) File "/usr/local/lib/python3.7/dist-packages/torch/nn/modules/module.py", line 1130, in _call_impl return forward_call(*input, **kwargs) TypeError: forward() takes 3 positional arguments but 6 were given

when I add mask in folder used render_cy, sometimes I get error face mask

but when I rewrite this cython to a python function like def sim_render_cy(vertices, new_colors, triangles, h, w): vertices_shape0 = vertices.shape[1] vis_colors = np.ones((1, vertices_shape0)) face_mask = render_texture(vertices, vis_colors, triangles, h, w, 1) new_image = render_texture(vertices, new_colors, triangles, h, w, 3) return face_mask, new_image

This error was resolved but it becomes very slow.

Hi, Thanks for sharing,

With face-down image, When I detect and align face, I have got issues: The landmark of face is wrong. Please help me check it