InsightFace_TF
InsightFace_TF copied to clipboard
auroua
Insight Face on TensorFlow
Hi, I found the implementation of accuracy just take the logits and then pass it into a softmax. ``` pred = tf.nn.softmax(logit) acc = tf.reduce_mean(tf.cast(tf.equal(tf.argmax(pred, axis=1), labels), dtype=tf.float32)) ``` However,...
Hello! I've found two types of performance issues in your program: - `batch()` should be called before `map()`. - `tf.Session` being defined repeatedly leads to incremental overhead. You can make...
Hello,I found some performance issues. **The first one** is in `train_nets.py`, [dataset = dataset.map(parse_function)](https://github.com/auroua/InsightFace_TF/blob/6ffe4296460bdfea56f91521db6d6412a89249d9/train_nets.py#L61) was called without **num_parallel_calls**. I think it will increase the efficiency of your program if you...
I use the same data train with insightface and facenet ,facenet turns out to be better ,so please give up insightface
epoch 0, total_step 300, total loss is 51.74 , inference loss is 37.28, weight deacy loss is 14.46, training accuracy is 0.000000, time 33.227 samples/sec epoch 0, total_step 320, total...
@auroua hi： the tricks as follows： in eval_data_reader.py “if flip == 1: img = np.fliplr(img) ” and in verification.py "embeddings = embeddings_list[0] + embeddings_list[1] embeddings = sklearn.preprocessing.normalize(embeddings) " so ,...
I don't have Baidu account. So, How could i can download pre-train model without account ?
Im trying to implement arcface loss on my custom model(building my own model arch) and custom dataset i want to know what is the label format (one hot encoded or...
I am embedding with `resnet_v1_50/E_BN2/Identity:0` I count the distance ``` dot = np.sum(np.multiply(em1, em2), axis=1) norm = np.linalg.norm(em1, axis=1) * np.linalg.norm(em2, axis=1) similarity = dot / norm ``` But get...
