Recall keeps zero

[Yuantian013]

Hi, I use the official raccoon_dataset-master data as my dataset, and train with the tiny yolo, but the current recall will keep in 0 even during 10 more epochs.I can't find out the problem. This is my json: { "model" : { "backend": "Tiny Yolo", "input_size": 200, "anchors": [1.94,2.51, 2.69,4.55, 3.62,5.55, 4.77,4.23, 5.31,5.78], "max_box_per_image": 10, "labels": ["raccoon"] },

"train": {
    "train_image_folder":   "/Users/tianyuan/Desktop/keras-yolo2-master/raccoon_dataset-master/images/",
    "train_annot_folder":   "/Users/tianyuan/Desktop/keras-yolo2-master/raccoon_dataset-master/annotations/",     

    "train_times":          1,
    "pretrained_weights":   "",
    "batch_size":           16,
    "learning_rate":        1e-4,
    "nb_epochs":            50,
    "warmup_epochs":        3,

    "object_scale":         5.0 ,
    "no_object_scale":      0.5,
    "coord_scale":          2.0,
    "class_scale":          1.0,

    "saved_weights_name":   "/Users/tianyuan/Desktop/keras-yolo2-master/test_weight.h5",
    "debug":                true
},

"valid": {
    "valid_image_folder":   "",
    "valid_annot_folder":   "",

    "valid_times":          1
}

}

[neney1234]

Same problem

[NPetsky]

I see from your config that you are in the warmup mode I think then recall can be zero (this is not the real training) see #117 #164 and #293 Set warmup_epochs to zero and pretrained_weights to those you have now saved (if you have not quit the warmup training) then you should start training and after some iterations recall should be non zero Hope I could help!

[letilessa]

Hi @Yuantian013 and @neney1234, did this answer work for you? I tried but I am still getting recall zero.