Hello

Are you using Multiscale Training of Data..also You have Pretrained Weights on VOC Data ...Below is the Image of Blood Smear :-

I Want to Detect The Purple Color and Red Color Cells...I have Done the Annotations ... I Only have 300 Images With Me with 15-20 Annotation in an Image...What do u Recommand ...

ValueError: Cannot feed value of shape (30,) for Tensor u'Placeholder_41:0', which has shape '(35,)'

Getting this error when I try to train two classes on tiny_yolo_raccoon.h5 pretrained weight when I train for one class its not a problem but when I do it for two this is the error I am getting. I want it for traffic sign detecting and classification with more than 10 classes can I do it also if anyone have weights for multi-class training please share it it would be much helpful.

Can anyone help. Thanks in advance.

Hi :) My goal is to understand how I can apply fine tuning to such neural network, and so I want to run your code and play with it ;) My problem is that deep learning is a new topic for my studies, so I have some problems to understand each part of your software. I have read the YOLO paper and I have an idea about how it works. I have prepared dataset, as you suggest in the section "Usage for python Code", but my first question is:

• Who generates annotations from images in VOC format? Are annotations only labels for images? What is their meaning?

Once that annotations are generated, I have to generate anchors. Does your gen_anchors script updates the raw about anchors in the conf.json file? Thank you so much for your response!

Hi @experiencor , i always have the same error appeared every time and i can't solve it the error is UnboundLocalError: local variable 'epoch_logs' referenced before assignment

how to solve it.

Can anybody help me in building YOLOv3 architecture in keras? here is the link I think its residual network https://github.com/pjreddie/darknet/blob/master/cfg/yolov3.cfg @experiencor

Hi, When I run python train.py -c config.json. The following error appears:

Using TensorFlow backend.
Traceback (most recent call last):
  File "train.py", line 140, in <module>
    _main_(args)
  File "train.py", line 76, in _main_
    config['model']['labels'])
  File "/home/julio_jcgc/yolo_tutorial/basic-yolo-keras/preprocessing.py", line 21, in parse_annotation
    tree = ET.parse(ann_dir + ann)
  File "/opt/bitnami/python/lib/python2.7/xml/etree/ElementTree.py", line 1182, in parse
    tree.parse(source, parser)
  File "/opt/bitnami/python/lib/python2.7/xml/etree/ElementTree.py", line 651, in parse
    parser = XMLParser(target=TreeBuilder())
  File "/opt/bitnami/python/lib/python2.7/xml/etree/ElementTree.py", line 1476, in __init__
    "No module named expat; use SimpleXMLTreeBuilder instead"
ImportError: No module named expat; use SimpleXMLTreeBuilder instead

Thank you.

I am trying to load the VOC YOLOv2 weights from the yolo website. I am working in the jupyter notebook provided in this repository. Here are the modified parameters:

LABELS = ['Person', 'Car', 'Bicycle', 'Bus', 'Motorbike', 'Train', 'Aeroplane', 'Chair', 'Bottle', 'Dining Table', 'Potted Plant', 'TV/Monitor', 'Sofa', 'Bird', 'Cat', 'Cow', 'Dog', 'Horse', 'Sheep']

IMAGE_H, IMAGE_W = 416, 416
GRID_H,  GRID_W  = 13 , 13
BOX              = 5
CLASS            = len(LABELS)
CLASS_WEIGHTS    = np.ones(CLASS, dtype='float32')
OBJ_THRESHOLD    = 0.3#0.5
NMS_THRESHOLD    = 0.3#0.45
ANCHORS          = [0.57273, 0.677385, 1.87446, 2.06253, 3.33843, 5.47434, 7.88282, 3.52778, 9.77052, 9.16828]

NO_OBJECT_SCALE  = 1.0
OBJECT_SCALE     = 5.0
COORD_SCALE      = 1.0
CLASS_SCALE      = 1.0

BATCH_SIZE       = 16
WARM_UP_BATCHES  = 0
TRUE_BOX_BUFFER  = 50

Here is the model summary:

Layer (type)                    Output Shape         Param #     Connected to                     
==================================================================================================
input_5 (InputLayer)            (None, 416, 416, 3)  0                                            
__________________________________________________________________________________________________
conv_0 (Conv2D)                 (None, 416, 416, 32) 864         input_5[0][0]                    
__________________________________________________________________________________________________
batch_norm_0 (BatchNormalizatio (None, 416, 416, 32) 128         conv_0[0][0]                     
__________________________________________________________________________________________________
leaky_re_lu_45 (LeakyReLU)      (None, 416, 416, 32) 0           batch_norm_0[0][0]               
__________________________________________________________________________________________________
max_pooling2d_11 (MaxPooling2D) (None, 208, 208, 32) 0           leaky_re_lu_45[0][0]             
__________________________________________________________________________________________________
conv_1 (Conv2D)                 (None, 208, 208, 64) 18432       max_pooling2d_11[0][0]           
__________________________________________________________________________________________________
batch_norm_1 (BatchNormalizatio (None, 208, 208, 64) 256         conv_1[0][0]                     
__________________________________________________________________________________________________
leaky_re_lu_46 (LeakyReLU)      (None, 208, 208, 64) 0           batch_norm_1[0][0]               
__________________________________________________________________________________________________
max_pooling2d_12 (MaxPooling2D) (None, 104, 104, 64) 0           leaky_re_lu_46[0][0]             
__________________________________________________________________________________________________
conv_2 (Conv2D)                 (None, 104, 104, 128 73728       max_pooling2d_12[0][0]           
__________________________________________________________________________________________________
batch_norm_2 (BatchNormalizatio (None, 104, 104, 128 512         conv_2[0][0]                     
__________________________________________________________________________________________________
leaky_re_lu_47 (LeakyReLU)      (None, 104, 104, 128 0           batch_norm_2[0][0]               
__________________________________________________________________________________________________
conv_3 (Conv2D)                 (None, 104, 104, 64) 8192        leaky_re_lu_47[0][0]             
__________________________________________________________________________________________________
batch_norm_3 (BatchNormalizatio (None, 104, 104, 64) 256         conv_3[0][0]                     
__________________________________________________________________________________________________
leaky_re_lu_48 (LeakyReLU)      (None, 104, 104, 64) 0           batch_norm_3[0][0]               
__________________________________________________________________________________________________
conv_4 (Conv2D)                 (None, 104, 104, 128 73728       leaky_re_lu_48[0][0]             
__________________________________________________________________________________________________
batch_norm_4 (BatchNormalizatio (None, 104, 104, 128 512         conv_4[0][0]                     
__________________________________________________________________________________________________
leaky_re_lu_49 (LeakyReLU)      (None, 104, 104, 128 0           batch_norm_4[0][0]               
__________________________________________________________________________________________________
max_pooling2d_13 (MaxPooling2D) (None, 52, 52, 128)  0           leaky_re_lu_49[0][0]             
__________________________________________________________________________________________________
conv_5 (Conv2D)                 (None, 52, 52, 256)  294912      max_pooling2d_13[0][0]           
__________________________________________________________________________________________________
batch_norm_5 (BatchNormalizatio (None, 52, 52, 256)  1024        conv_5[0][0]                     
__________________________________________________________________________________________________
leaky_re_lu_50 (LeakyReLU)      (None, 52, 52, 256)  0           batch_norm_5[0][0]               
__________________________________________________________________________________________________
conv_6 (Conv2D)                 (None, 52, 52, 128)  32768       leaky_re_lu_50[0][0]             
__________________________________________________________________________________________________
batch_norm_6 (BatchNormalizatio (None, 52, 52, 128)  512         conv_6[0][0]                     
__________________________________________________________________________________________________
leaky_re_lu_51 (LeakyReLU)      (None, 52, 52, 128)  0           batch_norm_6[0][0]               
__________________________________________________________________________________________________
conv_7 (Conv2D)                 (None, 52, 52, 256)  294912      leaky_re_lu_51[0][0]             
__________________________________________________________________________________________________
batch_norm_7 (BatchNormalizatio (None, 52, 52, 256)  1024        conv_7[0][0]                     
__________________________________________________________________________________________________
leaky_re_lu_52 (LeakyReLU)      (None, 52, 52, 256)  0           batch_norm_7[0][0]               
__________________________________________________________________________________________________
max_pooling2d_14 (MaxPooling2D) (None, 26, 26, 256)  0           leaky_re_lu_52[0][0]             
__________________________________________________________________________________________________
conv_8 (Conv2D)                 (None, 26, 26, 512)  1179648     max_pooling2d_14[0][0]           
__________________________________________________________________________________________________
batch_norm_8 (BatchNormalizatio (None, 26, 26, 512)  2048        conv_8[0][0]                     
__________________________________________________________________________________________________
leaky_re_lu_53 (LeakyReLU)      (None, 26, 26, 512)  0           batch_norm_8[0][0]               
__________________________________________________________________________________________________
conv_9 (Conv2D)                 (None, 26, 26, 256)  131072      leaky_re_lu_53[0][0]             
__________________________________________________________________________________________________
batch_norm_9 (BatchNormalizatio (None, 26, 26, 256)  1024        conv_9[0][0]                     
__________________________________________________________________________________________________
leaky_re_lu_54 (LeakyReLU)      (None, 26, 26, 256)  0           batch_norm_9[0][0]               
__________________________________________________________________________________________________
conv_10 (Conv2D)                (None, 26, 26, 512)  1179648     leaky_re_lu_54[0][0]             
__________________________________________________________________________________________________
batch_norm_10 (BatchNormalizati (None, 26, 26, 512)  2048        conv_10[0][0]                    
__________________________________________________________________________________________________
leaky_re_lu_55 (LeakyReLU)      (None, 26, 26, 512)  0           batch_norm_10[0][0]              
__________________________________________________________________________________________________
conv_11 (Conv2D)                (None, 26, 26, 256)  131072      leaky_re_lu_55[0][0]             
__________________________________________________________________________________________________
batch_norm_11 (BatchNormalizati (None, 26, 26, 256)  1024        conv_11[0][0]                    
__________________________________________________________________________________________________
leaky_re_lu_56 (LeakyReLU)      (None, 26, 26, 256)  0           batch_norm_11[0][0]              
__________________________________________________________________________________________________
conv_12 (Conv2D)                (None, 26, 26, 512)  1179648     leaky_re_lu_56[0][0]             
__________________________________________________________________________________________________
batch_norm_12 (BatchNormalizati (None, 26, 26, 512)  2048        conv_12[0][0]                    
__________________________________________________________________________________________________
leaky_re_lu_57 (LeakyReLU)      (None, 26, 26, 512)  0           batch_norm_12[0][0]              
__________________________________________________________________________________________________
max_pooling2d_15 (MaxPooling2D) (None, 13, 13, 512)  0           leaky_re_lu_57[0][0]             
__________________________________________________________________________________________________
conv_13 (Conv2D)                (None, 13, 13, 1024) 4718592     max_pooling2d_15[0][0]           
__________________________________________________________________________________________________
batch_norm_13 (BatchNormalizati (None, 13, 13, 1024) 4096        conv_13[0][0]                    
__________________________________________________________________________________________________
leaky_re_lu_58 (LeakyReLU)      (None, 13, 13, 1024) 0           batch_norm_13[0][0]              
__________________________________________________________________________________________________
conv_14 (Conv2D)                (None, 13, 13, 512)  524288      leaky_re_lu_58[0][0]             
__________________________________________________________________________________________________
batch_norm_14 (BatchNormalizati (None, 13, 13, 512)  2048        conv_14[0][0]                    
__________________________________________________________________________________________________
leaky_re_lu_59 (LeakyReLU)      (None, 13, 13, 512)  0           batch_norm_14[0][0]              
__________________________________________________________________________________________________
conv_15 (Conv2D)                (None, 13, 13, 1024) 4718592     leaky_re_lu_59[0][0]             
__________________________________________________________________________________________________
batch_norm_15 (BatchNormalizati (None, 13, 13, 1024) 4096        conv_15[0][0]                    
__________________________________________________________________________________________________
leaky_re_lu_60 (LeakyReLU)      (None, 13, 13, 1024) 0           batch_norm_15[0][0]              
__________________________________________________________________________________________________
conv_16 (Conv2D)                (None, 13, 13, 512)  524288      leaky_re_lu_60[0][0]             
__________________________________________________________________________________________________
batch_norm_16 (BatchNormalizati (None, 13, 13, 512)  2048        conv_16[0][0]                    
__________________________________________________________________________________________________
leaky_re_lu_61 (LeakyReLU)      (None, 13, 13, 512)  0           batch_norm_16[0][0]              
__________________________________________________________________________________________________
conv_17 (Conv2D)                (None, 13, 13, 1024) 4718592     leaky_re_lu_61[0][0]             
__________________________________________________________________________________________________
batch_norm_17 (BatchNormalizati (None, 13, 13, 1024) 4096        conv_17[0][0]                    
__________________________________________________________________________________________________
leaky_re_lu_62 (LeakyReLU)      (None, 13, 13, 1024) 0           batch_norm_17[0][0]              
__________________________________________________________________________________________________
conv_18 (Conv2D)                (None, 13, 13, 1024) 9437184     leaky_re_lu_62[0][0]             
__________________________________________________________________________________________________
batch_norm_18 (BatchNormalizati (None, 13, 13, 1024) 4096        conv_18[0][0]                    
__________________________________________________________________________________________________
conv_20 (Conv2D)                (None, 26, 26, 64)   32768       leaky_re_lu_57[0][0]             
__________________________________________________________________________________________________
leaky_re_lu_63 (LeakyReLU)      (None, 13, 13, 1024) 0           batch_norm_18[0][0]              
__________________________________________________________________________________________________
batch_norm_20 (BatchNormalizati (None, 26, 26, 64)   256         conv_20[0][0]                    
__________________________________________________________________________________________________
conv_19 (Conv2D)                (None, 13, 13, 1024) 9437184     leaky_re_lu_63[0][0]             
__________________________________________________________________________________________________
leaky_re_lu_65 (LeakyReLU)      (None, 26, 26, 64)   0           batch_norm_20[0][0]              
__________________________________________________________________________________________________
batch_norm_19 (BatchNormalizati (None, 13, 13, 1024) 4096        conv_19[0][0]                    
__________________________________________________________________________________________________
lambda_4 (Lambda)               (None, 13, 13, 256)  0           leaky_re_lu_65[0][0]             
__________________________________________________________________________________________________
leaky_re_lu_64 (LeakyReLU)      (None, 13, 13, 1024) 0           batch_norm_19[0][0]              
__________________________________________________________________________________________________
concatenate_3 (Concatenate)     (None, 13, 13, 1280) 0           lambda_4[0][0]                   
                                                                 leaky_re_lu_64[0][0]             
__________________________________________________________________________________________________
conv_21 (Conv2D)                (None, 13, 13, 1024) 11796480    concatenate_3[0][0]              
__________________________________________________________________________________________________
batch_norm_21 (BatchNormalizati (None, 13, 13, 1024) 4096        conv_21[0][0]                    
__________________________________________________________________________________________________
leaky_re_lu_66 (LeakyReLU)      (None, 13, 13, 1024) 0           batch_norm_21[0][0]              
__________________________________________________________________________________________________
conv2d_3 (Conv2D)               (None, 13, 13, 120)  123000      leaky_re_lu_66[0][0]             
__________________________________________________________________________________________________
reshape_3 (Reshape)             (None, 13, 13, 5, 24 0           conv2d_3[0][0]                   
__________________________________________________________________________________________________
input_4 (InputLayer)            (None, 1, 1, 1, 50,  0                                            
__________________________________________________________________________________________________
lambda_5 (Lambda)               (None, 13, 13, 5, 24 0           reshape_3[0][0]                  
                                                                 input_4[0][0]                    
==================================================================================================
Total params: 50,670,936
Trainable params: 50,650,264
Non-trainable params: 20,672
__________________________________________________________________________________________________

I "succesfully" read the weights (succesfully as in no errors):

weight_reader = WeightReader('yolov2-voc.weights')

for index in range(conv_count):
    conv_layer = model.get_layer('conv_%i' % index)
    norm_layer = model.get_layer('batch_norm_%i' % index)
    
    size = np.prod(norm_layer.get_weights()[0].shape) # get product of shape (total values)
    
    # read sizes
    beta  = weight_reader.read(size)
    gamma = weight_reader.read(size)
    mean  = weight_reader.read(size)
    var   = weight_reader.read(size)
    
    norm_layer.set_weights([gamma, beta, mean, var])
    
    if len(conv_layer.get_weights()) > 1: 
        bias   = weight_reader.read(np.prod(conv_layer.get_weights()[1].shape)) 
        kernel = weight_reader.read(np.prod(conv_layer.get_weights()[0].shape))
        kernel = kernel.reshape(list(reversed(conv_layer.get_weights()[0].shape))) 
        kernel = kernel.transpose([2, 3, 1, 0])
        conv_layer.set_weights([kernel, bias])
    else:
        kernel = weight_reader.read(np.prod(conv_layer.get_weights()[0].shape))
        kernel = kernel.reshape(list(reversed(conv_layer.get_weights()[0].shape)))
        kernel = kernel.transpose([2, 3, 1, 0])
        conv_layer.set_weights([kernel])

When I try to run the following code, no boxes are created:

img = cv2.imread("dog-cycle-car.png")
img = cv2.resize(img, (416, 416)) # resize to the input dimension
img = img / 255
img = img[..., ::-1] # .transpose((2, 0, 1))  # BGR -> RGB | H X W C -> C X H X W 
img_input = np.array([img])

dummy_array = np.zeros((1, 1, 1, 1, TRUE_BOX_BUFFER, 4))

test_prediction = model.predict([img_input, dumby_array])
boxes = decode_netout(test_prediction[0],
                      obj_threshold=OBJ_THRESHOLD,
                      nms_threshold=NMS_THRESHOLD,
                      anchors=ANCHORS, 
                      nb_class=CLASS)
img = draw_boxes(img, boxes, labels=LABELS)
img.shape
plt.imshow(img)
boxes

I assume that there must be an issue with how I am loading the weights because the summary of the model looks the same and that is the only thing that I have modified in the notebook. I would be super grateful if someone could shine some light on to why this may be happening and what I am doing wrong.

Edit

If it would be helpful to see my jupyter notebook just ask for a link.

Hi , i want to trained the full-yolo with one GTX 1080 on raccoon. Firstly , please see my config.json :

{
    "model" : {
        "architecture":         "Full Yolo",
        "input_size":           416,
        "anchors":            [0.57273, 0.677385, 1.87446, 2.06253, 3.33843, 5.47434, 7.88282, 3.52778, 9.77052, 9.16828],
        "max_box_per_image":    20,        
        "labels":               ["raccoon"]

    },

    "train": {
        "train_image_folder":   "/home/mm/Detection-keras/basic-yolo-keras/raccoon_dataset/images/train/",
        "train_annot_folder":   "/home/mm/Detection-keras/basic-yolo-keras/raccoon_dataset/annotations/train/",     
          
        "train_times":          10,
        "pretrained_weights":   "",
        "batch_size":           12,
        "learning_rate":        1e-4,
        "nb_epoch":             50,
        "warmup_epochs":        3,

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

        "saved_weights_name":   "full_yolo_raccoon.h5",
        "debug":                true
    },

    "valid": {
        "valid_image_folder":   "/home/mm/Detection-keras/basic-yolo-keras/raccoon_dataset/images/val/",
        "valid_annot_folder":   "/home/mm/Detection-keras/basic-yolo-keras/raccoon_dataset/annotations/val/",

        "valid_times":          1
    }
}

Now i train , Epoch 1 : current recalls are above 98 % http://uupload.ir/files/jv2c_screenshot_from_2018-03-14_21-54-08.png

Epoch 1 : couple of iterations : recall goes to zeros. http://uupload.ir/files/pxfl_screenshot_from_2018-03-14_21-54-27.png

Epoch 3 , End of training , Still recall=0 http://uupload.ir/files/i2qp_screenshot_from_2018-03-14_21-57-35.png

when i tested on raccoon images , no object found . http://uupload.ir/files/sgms_screenshot_from_2018-03-14_22-21-59.png

Why? where are the problems ?

Hi, it seems that your script always assume that the image has 3 channels:

input_image = Input(shape=(self.input_size, self.input_size, 3))

It would be nice if you change the script that it also supports single channgel images.

Thanks for the nice work!

I've done some changes to train the model on my own data set.(4 classes). Model architecture is formed and while training epoch 1, the following error is coming:

Epoch 1/100000 Traceback (most recent call last): File "train.py", line 137, in main(args) File "train.py", line 133, in main debug = config['train']['debug']) File "/home/bhanu/Yolo-Keras/frontend.py", line 447, in train max_queue_size = 8) File "/usr/local/lib/python3.5/dist-packages/Keras-2.1.1-py3.5.egg/keras/legacy/interfaces.py", line 87, in wrapper File "/usr/local/lib/python3.5/dist-packages/Keras-2.1.1-py3.5.egg/keras/engine/training.py", line 2114, in fit_generator File "/usr/local/lib/python3.5/dist-packages/Keras-2.1.1-py3.5.egg/keras/engine/training.py", line 1826, in train_on_batch File "/usr/local/lib/python3.5/dist-packages/Keras-2.1.1-py3.5.egg/keras/engine/training.py", line 1411, in _standardize_user_data File "/usr/local/lib/python3.5/dist-packages/Keras-2.1.1-py3.5.egg/keras/engine/training.py", line 153, in _standardize_input_data ValueError: Error when checking target: expected lambda_2 to have shape (None, 11, 11, 5, 9) but got array with shape (1, 11, 11, 5, 6)

Please help me with this.

Hi. I have a problem when training with Raccoon dataset. I dont know why it always stop when I run after a few minutes. I re-train many time but the result no change.

This is my config:

{
    "model" : {
        "architecture":         "Tiny Yolo",
        "input_size":           416,
        "anchors":              [0.57273, 0.677385, 1.87446, 2.06253, 3.33843, 5.47434, 7.88282, 3.52778, 9.77052, 9.16828],
        "max_box_per_image":    10,        
        "labels":               ["raccoon"]
    },

    "train": {
        "train_image_folder":   "../raccoon_dataset/train/images/",
        "train_annot_folder":   "../raccoon_dataset/train/annotations/",     
          
        "train_times":          10,
        "pretrained_weights":   "",
        "batch_size":           16,
        "learning_rate":        1e-4,
        "nb_epoch":             50,
        "warmup_epochs":        0,

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

        "saved_weights_name":   "tiny_yolo_raccoon_save.h5",
        "debug":                false
    },

    "valid": {
        "valid_image_folder":   "../raccoon_dataset/valid/images/",
        "valid_annot_folder":   "../raccoon_dataset/valid/annotations/",
        "valid_times":          1
    }
}

Epoch 1/50
 10/100 [==>...........................] - ETA: 1:11 - loss: 11.5398Epoch 00001: val_loss improved from inf to 5.79008, saving model to tiny_yolo_raccoon_save.h5
 10/100 [==>...........................] - ETA: 1:22 - loss: 11.5398 - val_loss: 0.0000e+00Epoch 2/50
 10/100 [==>...........................] - ETA: 44s - loss: 9.3070Epoch 00002: val_loss improved from 5.79008 to 5.75633, saving model to tiny_yolo_raccoon_save.h5
 10/100 [==>...........................] - ETA: 55s - loss: 9.3070 - val_loss: 0.0000e+00Epoch 3/50
 10/100 [==>...........................] - ETA: 44s - loss: 6.9476Epoch 00003: val_loss improved from 5.75633 to 4.91244, saving model to tiny_yolo_raccoon_save.h5
 10/100 [==>...........................] - ETA: 55s - loss: 6.9476 - val_loss: 0.0000e+00Epoch 4/50
 10/100 [==>...........................] - ETA: 44s - loss: 4.7331Epoch 00004: val_loss improved from 4.91244 to 4.88484, saving model to tiny_yolo_raccoon_save.h5
 10/100 [==>...........................] - ETA: 55s - loss: 4.7331 - val_loss: 0.0000e+00Epoch 5/50
 10/100 [==>...........................] - ETA: 44s - loss: 3.8645Epoch 00005: val_loss did not improve
 10/100 [==>...........................] - ETA: 53s - loss: 3.8645 - val_loss: 0.0000e+00Epoch 6/50
 10/100 [==>...........................] - ETA: 44s - loss: 3.8343Epoch 00006: val_loss did not improve
 10/100 [==>...........................] - ETA: 52s - loss: 3.8343 - val_loss: 0.0000e+00Epoch 7/50
 10/100 [==>...........................] - ETA: 44s - loss: 3.4981Epoch 00007: val_loss did not improve
 10/100 [==>...........................] - ETA: 53s - loss: 3.4981 - val_loss: 0.0000e+00Epoch 00007: early stopping

It train in to few minute ??? Too fast !!!!!. Are there something wrong in here, the val_loss is 0.0000+e00, but there are no box found when I predict ? And I have question, why the progress bar only run to 10/100 each epoch ? I never seen it run to 100/100 ? Thank you so much, hope anyone reply, I tried many many times but the result seem not good :(

Hi, I have been trying for a long time to train my own data set with this notebook but I have faced many issues and solved some of them Firsy My Data Set is as follows images folder ----> with training images of size 640x512 labels folder ---> text files for each image above with following data 0 311.489379882812 204.399459838867 36.0547180175781 24.1059265136719 i-e class label, x , y, w, h, (these values are not normalized.

I have modified preprocessinmg.py to output the dataset as required by the notebook but still I get the error

c:\Users\Usman\anaconda3\envs\tf-gpu\lib\site-packages\numpy\core\fromnumeric.py:43 _wrapit
    result = getattr(asarray(obj), method)(*args, **kwds)

ValueError: cannot reshape array of size 10 into shape (1,1,1,1,2)

and i cannt seen to find where my error is

Ihave boon stuck at this for like 4 doays now please help

Bumps tensorflow-gpu from 1.3 to 2.9.3.

Dependabot will resolve any conflicts with this PR as long as you don't alter it yourself. You can also trigger a rebase manually by commenting @dependabot rebase.

tf.Print is deprecated so I get a Tensorflow warning about replacing it with tf.print, which unfortunately doesn't work either for some reason. After a bit of online research I think I get how tf.print works on Tensorflow 2, but I'm still confused about its function on Tensorflow 1. I run the Yolo_Step_by_Step.ipynb with Tensorflow 1.x. Has anyone been able to make these prints work?

I am using the notebook on colab and I want to run it with TF2. However, I come across this error on calling model.fit_generator:

NotFoundError:` 2 root error(s) found.
  (0) Not found: Resource localhost/loss/lambda_3_loss/Variable/N10tensorflow3VarE does not exist.
	 [[{{node loss/lambda_3_loss/AssignAddVariableOp}}]]
	 [[Func/training/Adam/gradients/gradients/norm_6_1/cond_grad/StatelessIf/then/_1694/input/_4425/_2667]]
  (1) Not found: Resource localhost/loss/lambda_3_loss/Variable/N10tensorflow3VarE does not exist.
	 [[{{node loss/lambda_3_loss/AssignAddVariableOp}}]]
0 successful operations.
0 derived errors ignored.

The problem seems to be caused by the custom loss function, since I tried using a simple dummy loss function with no errors.

The changes I've made (to no avail) are these 2:

1. to include the lines from tensorflow.python.framework.ops import disable_eager_execution disable_eager_execution() before creating the model. Before adding these lines I came across this error:

TypeError: Cannot convert a symbolic Keras input/output to a numpy array.
This error may indicate that you're trying to pass a symbolic value to a NumPy call,
which is not supported.
Or, you may be trying to pass Keras symbolic inputs/outputs to a TF API that does not register dispatching,
preventing Keras from automatically converting the API call to a lambda layer in the Functional Model.

2. to use tensorflow.keras instead of keras, after suggestions from similar github issues and stackoverflow posts.

I am also adding the custom_loss code to include a few changes I made to use TF2 instead of TF1 (basically some tf.compat.v1.* additions).

def custom_loss(y_true, y_pred):
    mask_shape = tf.shape(y_true)[:4]
    
    cell_x = tf.compat.v1.to_float(tf.reshape(tf.tile(tf.range(GRID_W), [GRID_H]), (1, GRID_H, GRID_W, 1, 1)))
    cell_y = tf.transpose(cell_x, (0,2,1,3,4))

    cell_grid = tf.tile(tf.concat([cell_x,cell_y], -1), [BATCH_SIZE, 1, 1, 5, 1])
    
    coord_mask = tf.zeros(mask_shape)
    conf_mask  = tf.zeros(mask_shape)
    class_mask = tf.zeros(mask_shape)
    
    seen = tf.Variable(0.)
    total_recall = tf.Variable(0.)
    
    """
    Adjust prediction
    """
    ### adjust x and y      
    pred_box_xy = tf.sigmoid(y_pred[..., :2]) + cell_grid
    
    ### adjust w and h
    pred_box_wh = tf.exp(y_pred[..., 2:4]) * np.reshape(ANCHORS, [1,1,1,BOX,2])
    
    ### adjust confidence
    pred_box_conf = tf.sigmoid(y_pred[..., 4])
    
    ### adjust class probabilities
    pred_box_class = y_pred[..., 5:]
    
    """
    Adjust ground truth
    """
    ### adjust x and y
    true_box_xy = y_true[..., 0:2] # relative position to the containing cell
    
    ### adjust w and h
    true_box_wh = y_true[..., 2:4] # number of cells accross, horizontally and vertically
    
    ### adjust confidence
    true_wh_half = true_box_wh / 2.
    true_mins    = true_box_xy - true_wh_half
    true_maxes   = true_box_xy + true_wh_half
    
    pred_wh_half = pred_box_wh / 2.
    pred_mins    = pred_box_xy - pred_wh_half
    pred_maxes   = pred_box_xy + pred_wh_half       
    
    intersect_mins  = tf.maximum(pred_mins,  true_mins)
    intersect_maxes = tf.minimum(pred_maxes, true_maxes)
    intersect_wh    = tf.maximum(intersect_maxes - intersect_mins, 0.)
    intersect_areas = intersect_wh[..., 0] * intersect_wh[..., 1]
    
    true_areas = true_box_wh[..., 0] * true_box_wh[..., 1]
    pred_areas = pred_box_wh[..., 0] * pred_box_wh[..., 1]

    union_areas = pred_areas + true_areas - intersect_areas
    iou_scores  = tf.truediv(intersect_areas, union_areas)
    
    true_box_conf = iou_scores * y_true[..., 4]
    
    ### adjust class probabilities
    true_box_class = tf.argmax(y_true[..., 5:], -1)
    
    """
    Determine the masks
    """
    ### coordinate mask: simply the position of the ground truth boxes (the predictors)
    coord_mask = tf.expand_dims(y_true[..., 4], axis=-1) * COORD_SCALE
    
    ### confidence mask: penelize predictors + penalize boxes with low IOU
    # penalize the confidence of the boxes, which have IOU with some ground truth box < 0.6
    true_xy = true_boxes[..., 0:2]
    true_wh = true_boxes[..., 2:4]
    
    true_wh_half = true_wh / 2.
    true_mins    = true_xy - true_wh_half
    true_maxes   = true_xy + true_wh_half
    
    pred_xy = tf.expand_dims(pred_box_xy, 4)
    pred_wh = tf.expand_dims(pred_box_wh, 4)
    
    pred_wh_half = pred_wh / 2.
    pred_mins    = pred_xy - pred_wh_half
    pred_maxes   = pred_xy + pred_wh_half    
    
    intersect_mins  = tf.maximum(pred_mins,  true_mins)
    intersect_maxes = tf.minimum(pred_maxes, true_maxes)
    intersect_wh    = tf.maximum(intersect_maxes - intersect_mins, 0.)
    intersect_areas = intersect_wh[..., 0] * intersect_wh[..., 1]
    
    true_areas = true_wh[..., 0] * true_wh[..., 1]
    pred_areas = pred_wh[..., 0] * pred_wh[..., 1]

    union_areas = pred_areas + true_areas - intersect_areas
    iou_scores  = tf.truediv(intersect_areas, union_areas)

    best_ious = tf.reduce_max(iou_scores, axis=4)
    conf_mask = conf_mask + tf.compat.v1.to_float(best_ious < 0.6) * (1 - y_true[..., 4]) * NO_OBJECT_SCALE
    
    # penalize the confidence of the boxes, which are reponsible for corresponding ground truth box
    conf_mask = conf_mask + y_true[..., 4] * OBJECT_SCALE
    
    ### class mask: simply the position of the ground truth boxes (the predictors)
    class_mask = y_true[..., 4] * tf.gather(CLASS_WEIGHTS, true_box_class) * CLASS_SCALE       
    
    """
    Warm-up training
    """
    no_boxes_mask = tf.compat.v1.to_float(coord_mask < COORD_SCALE/2.)
    seen = tf.compat.v1.assign_add(seen, 1.)
    
    true_box_xy, true_box_wh, coord_mask = tf.cond(tf.less(seen, WARM_UP_BATCHES), 
                          lambda: [true_box_xy + (0.5 + cell_grid) * no_boxes_mask, 
                                   true_box_wh + tf.ones_like(true_box_wh) * np.reshape(ANCHORS, [1,1,1,BOX,2]) * no_boxes_mask, 
                                   tf.ones_like(coord_mask)],
                          lambda: [true_box_xy, 
                                   true_box_wh,
                                   coord_mask])
    
    """
    Finalize the loss
    """
    nb_coord_box = tf.reduce_sum(tf.compat.v1.to_float(coord_mask > 0.0))
    nb_conf_box  = tf.reduce_sum(tf.compat.v1.to_float(conf_mask  > 0.0))
    nb_class_box = tf.reduce_sum(tf.compat.v1.to_float(class_mask > 0.0))
    
    loss_xy    = tf.reduce_sum(tf.square(true_box_xy-pred_box_xy)     * coord_mask) / (nb_coord_box + 1e-6) / 2.
    loss_wh    = tf.reduce_sum(tf.square(true_box_wh-pred_box_wh)     * coord_mask) / (nb_coord_box + 1e-6) / 2.
    loss_conf  = tf.reduce_sum(tf.square(true_box_conf-pred_box_conf) * conf_mask)  / (nb_conf_box  + 1e-6) / 2.
    loss_class = tf.nn.sparse_softmax_cross_entropy_with_logits(labels=true_box_class, logits=pred_box_class)
    loss_class = tf.reduce_sum(loss_class * class_mask) / (nb_class_box + 1e-6)
    
    loss = loss_xy + loss_wh + loss_conf + loss_class
    
    nb_true_box = tf.reduce_sum(y_true[..., 4])
    nb_pred_box = tf.reduce_sum(tf.compat.v1.to_float(true_box_conf > 0.5) * tf.compat.v1.to_float(pred_box_conf > 0.3))

    """
    Debugging code
    """    
    current_recall = nb_pred_box/(nb_true_box + 1e-6)
    total_recall = tf.compat.v1.assign_add(total_recall, current_recall) 

    loss = tf.compat.v1.Print(loss, [tf.zeros((1))], message='Dummy Line \t', summarize=1000)
    loss = tf.compat.v1.Print(loss, [loss_xy], message='Loss XY \t', summarize=1000)
    loss = tf.compat.v1.Print(loss, [loss_wh], message='Loss WH \t', summarize=1000)
    loss = tf.compat.v1.Print(loss, [loss_conf], message='Loss Conf \t', summarize=1000)
    loss = tf.compat.v1.Print(loss, [loss_class], message='Loss Class \t', summarize=1000)
    loss = tf.compat.v1.Print(loss, [loss], message='Total Loss \t', summarize=1000)
    loss = tf.compat.v1.Print(loss, [current_recall], message='Current Recall \t', summarize=1000)
    loss = tf.compat.v1.Print(loss, [total_recall/seen], message='Average Recall \t', summarize=1000)
    
    return loss