Newer
Older
" keras.callbacks.ModelCheckpoint(filepath=\"unet_segmentation.keras\", save_best_only=True, monitor=\"val_loss\"),\n",
" tf.keras.callbacks.TensorBoard(logdir, histogram_freq=1)\n",
"] \n",
"\n",
"history = model.fit(train_input_imgs, train_targets,\n",
" epochs=50,\n",
" callbacks=callbacks,\n",
" batch_size=64,\n",
" validation_data=(val_input_imgs, val_targets))"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"Let’s display our training and validation loss:"
]
},
{
"cell_type": "code",
3024
3025
3026
3027
3028
3029
3030
3031
3032
3033
3034
3035
3036
3037
3038
3039
3040
3041
3042
3043
3044
3045
3046
3047
3048
3049
"outputs": [
{
"data": {
"image/png": "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\n",
"text/plain": [
"<Figure size 432x288 with 1 Axes>"
]
},
"metadata": {
"needs_background": "light"
},
"output_type": "display_data"
},
{
"data": {
"image/png": "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\n",
"text/plain": [
"<Figure size 432x288 with 1 Axes>"
]
},
"metadata": {
"needs_background": "light"
},
"output_type": "display_data"
}
],
3050
3051
3052
3053
3054
3055
3056
3057
3058
3059
3060
3061
3062
3063
3064
3065
3066
3067
3068
3069
3070
3071
3072
3073
3074
"source": [
"plt.plot(history.history['accuracy'])\n",
"plt.plot(history.history['val_accuracy'])\n",
"plt.title('model accuracy')\n",
"plt.ylabel('accuracy')\n",
"plt.xlabel('epoch')\n",
"plt.legend(['train', 'valid'], loc='lower right')\n",
"plt.show()\n",
"plt.plot(history.history['loss'])\n",
"plt.plot(history.history['val_loss'])\n",
"plt.title('model loss')\n",
"plt.ylabel('loss')\n",
"plt.xlabel('epoch')\n",
"plt.legend(['train', 'valid'], loc='upper right')\n",
"plt.show()"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"You can see that we start overfitting midway, around epoch 25. Let’s reload our best\n",
"performing model according to the validation loss, and demonstrate how to use it to\n",
"predict a segmentation mask"
]
"outputs": [
{
"data": {
"image/png": "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\n",
"text/plain": [
"<Figure size 432x288 with 1 Axes>"
]
},
"metadata": {
"needs_background": "light"
},
"output_type": "display_data"
}
],
"source": [
"from tensorflow.keras.utils import array_to_img\n",
"model = keras.models.load_model(\"unet_segmentation.keras\")\n",
3097
3098
3099
3100
3101
3102
3103
3104
3105
3106
3107
3108
3109
3110
3111
3112
3113
3114
3115
3116
3117
3118
3119
"i = 4\n",
"test_image = val_input_imgs[i]\n",
"plt.axis(\"off\")\n",
"plt.imshow(array_to_img(test_image))\n",
"mask = model.predict(np.expand_dims(test_image, 0))[0]\n",
"\n",
"# Utility to display a model’s prediction\n",
"def display_mask(pred):\n",
" mask = np.argmax(pred, axis=-1)\n",
" mask *= 127\n",
" plt.axis(\"off\")\n",
" plt.imshow(mask)\n",
" \n",
"display_mask(mask)"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"There are a couple of small artifacts in our predicted mask, caused by geometric shapes\n",
"in the foreground and background. Nevertheless, our model appears to work nicely."
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"# Part IV : Object Detection with Yolo"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"The section follows the [blog](https://machinelearningmastery.com/how-to-perform-object-detection-with-yolov3-in-keras/). "
3135
3136
3137
3138
3139
3140
3141
3142
3143
3144
3145
3146
3147
3148
3149
3150
3151
3152
3153
3154
3155
3156
3157
3158
3159
3160
3161
3162
3163
3164
3165
3166
3167
{
"cell_type": "markdown",
"metadata": {},
"source": [
"Object detection is a computer vision task that involves both localizing one or more objects within an image and classifying each object in the image.\n",
"\n",
"It is a challenging computer vision task that requires both successful object localization in order to locate and draw a bounding box around each object in an image, and object classification to predict the correct class of object that was localized.\n",
"\n",
"The “You Only Look Once,” or YOLO, family of models are a series of end-to-end deep learning models designed for fast object detection, developed by Joseph Redmon, et al. and first described in the 2015 paper titled [You Only Look Once: Unified, Real-Time Object Detection](https://arxiv.org/abs/1506.02640).\n",
"\n",
"The approach involves a single deep convolutional neural network (originally a version of GoogLeNet, later updated and called DarkNet based on VGG) that splits the input into a grid of cells and each cell directly predicts a bounding box and object classification. The result is a large number of candidate bounding boxes that are consolidated into a final prediction by a post-processing step.\n",
"\n",
"There are three main variations of the approach, at the time of writing; they are YOLOv1, YOLOv2, and YOLOv3. The first version proposed the general architecture, whereas the second version refined the design and made use of predefined anchor boxes to improve bounding box proposal, and version three further refined the model architecture and training process.\n",
"\n",
"Although the accuracy of the models is close but not as good as Region-Based Convolutional Neural Networks (R-CNNs), they are popular for object detection because of their detection speed, often demonstrated in real-time on video or with camera feed input.\n",
"\n",
"A single neural network predicts bounding boxes and class probabilities directly from full images in one evaluation. Since the whole detection pipeline is a single network, it can be optimized end-to-end directly on detection performance."
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"### Object Detection With YOLOv3\n",
"\n",
"The keras-yolo3 project provides a lot of capability for using [YOLOv3 models](https://github.com/experiencor/keras-yolo3), including object detection, transfer learning, and training new models from scratch.\n",
"\n",
"In this section, we will use a pre-trained model to perform object detection on an unseen photograph. This capability is available in a single Python file in the repository called [yolo3_one_file_to_detect_them_all.py](https://raw.githubusercontent.com/experiencor/keras-yolo3/master/yolo3_one_file_to_detect_them_all.py) that has about 435 lines. This script is, in fact, a program that will use pre-trained weights to prepare a model and use that model to perform object detection and output a model. It also depends upon OpenCV.\n",
"\n",
"Instead of using this program directly, we will reuse elements from this program and develop our own scripts to first prepare and save a Keras YOLOv3 model, and then load the model to make a prediction for a new photograph."
]
},
{
"source": [
"### Create and Save Model\n",
"\n",
"The first step is to download the pre-trained model weights.\n",
"\n",
"These were trained using the DarkNet code base on the MSCOCO dataset. Download the model weights and place them into your current working directory with the filename `yolov3.weights`. It is a large file and may take a moment to download depending on the speed of your internet connection.\n",
"\n",
"[YOLOv3 Pre-trained Model Weights (yolov3.weights) (237 MB)](https://pjreddie.com/media/files/yolov3.weights)\n",
"\n",
"\n",
"Next, we need to define a Keras model that has the right number and type of layers to match the downloaded model weights. The model architecture is called a _DarkNet_ and was originally loosely based on the VGG-16 model.\n",
"\n",
"The `yolo3_one_file_to_detect_them_all.py` script provides the `make_yolov3_model()` function to create the model for us, and the helper function `_conv_block()` that is used to create blocks of layers. These two functions can be copied directly from the script.\n",
"\n",
"We can now define the Keras model for YOLOv3."
]
"source": [
"# define the model\n",
"model = make_yolov3_model()"
]
3198
3199
3200
3201
3202
3203
3204
3205
3206
3207
3208
3209
3210
3211
3212
3213
3214
3215
3216
3217
3218
3219
3220
3221
3222
3223
3224
3225
3226
3227
3228
3229
3230
3231
3232
3233
3234
3235
3236
3237
3238
3239
3240
3241
3242
3243
3244
3245
3246
3247
3248
3249
3250
3251
"source": [
"Next, we need to load the model weights. The model weights are stored in whatever format that was used by \n",
"_DarkNet_. Rather than trying to decode the file manually, we can use the `WeightReader` class provided in the script.\n",
"\n",
"To use the `WeightReader`, it is instantiated with the path to our weights file (e.g. `yolov3.weights`). This will parse the file \n",
"and load the model weights into memory in a format that we can set into our Keras model."
]
},
{
"cell_type": "raw",
"metadata": {},
"source": [
"# load the model weights\n",
"weight_reader = WeightReader('yolov3.weights')"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"We can then call the `load_weights()` function of the `WeightReader` instance, passing in our defined Keras model to set the weights into the layers."
]
},
{
"cell_type": "raw",
"metadata": {},
"source": [
"# set the model weights into the model\n",
"weight_reader.load_weights(model)"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"That’s it; we now have a _YOLOv3_ model for use.\n",
"\n",
"We can save this model to a Keras compatible `.h5` model file ready for later use."
]
},
{
"cell_type": "raw",
"metadata": {},
"source": [
"# save the model to file\n",
"model.save('model.h5')"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"We can tie all of this together; the complete code example including functions copied directly from the `yolo3_one_file_to_detect_them_all.py` script is listed below."
]
3259
3260
3261
3262
3263
3264
3265
3266
3267
3268
3269
3270
3271
3272
3273
3274
3275
3276
3277
3278
3279
3280
3281
3282
3283
3284
3285
3286
3287
3288
3289
3290
3291
3292
3293
3294
3295
3296
3297
3298
3299
3300
3301
3302
3303
3304
3305
3306
3307
3308
3309
3310
3311
3312
3313
3314
3315
3316
3317
3318
3319
3320
3321
3322
3323
3324
3325
3326
3327
3328
3329
3330
3331
3332
3333
3334
3335
3336
3337
3338
3339
3340
3341
3342
3343
3344
3345
3346
3347
3348
3349
3350
3351
3352
3353
3354
3355
3356
3357
3358
3359
3360
3361
3362
3363
3364
3365
3366
3367
3368
"name": "stdout",
"output_type": "stream",
"text": [
"loading weights of convolution #0\n",
"loading weights of convolution #1\n",
"loading weights of convolution #2\n",
"loading weights of convolution #3\n",
"no convolution #4\n",
"loading weights of convolution #5\n",
"loading weights of convolution #6\n",
"loading weights of convolution #7\n",
"no convolution #8\n",
"loading weights of convolution #9\n",
"loading weights of convolution #10\n",
"no convolution #11\n",
"loading weights of convolution #12\n",
"loading weights of convolution #13\n",
"loading weights of convolution #14\n",
"no convolution #15\n",
"loading weights of convolution #16\n",
"loading weights of convolution #17\n",
"no convolution #18\n",
"loading weights of convolution #19\n",
"loading weights of convolution #20\n",
"no convolution #21\n",
"loading weights of convolution #22\n",
"loading weights of convolution #23\n",
"no convolution #24\n",
"loading weights of convolution #25\n",
"loading weights of convolution #26\n",
"no convolution #27\n",
"loading weights of convolution #28\n",
"loading weights of convolution #29\n",
"no convolution #30\n",
"loading weights of convolution #31\n",
"loading weights of convolution #32\n",
"no convolution #33\n",
"loading weights of convolution #34\n",
"loading weights of convolution #35\n",
"no convolution #36\n",
"loading weights of convolution #37\n",
"loading weights of convolution #38\n",
"loading weights of convolution #39\n",
"no convolution #40\n",
"loading weights of convolution #41\n",
"loading weights of convolution #42\n",
"no convolution #43\n",
"loading weights of convolution #44\n",
"loading weights of convolution #45\n",
"no convolution #46\n",
"loading weights of convolution #47\n",
"loading weights of convolution #48\n",
"no convolution #49\n",
"loading weights of convolution #50\n",
"loading weights of convolution #51\n",
"no convolution #52\n",
"loading weights of convolution #53\n",
"loading weights of convolution #54\n",
"no convolution #55\n",
"loading weights of convolution #56\n",
"loading weights of convolution #57\n",
"no convolution #58\n",
"loading weights of convolution #59\n",
"loading weights of convolution #60\n",
"no convolution #61\n",
"loading weights of convolution #62\n",
"loading weights of convolution #63\n",
"loading weights of convolution #64\n",
"no convolution #65\n",
"loading weights of convolution #66\n",
"loading weights of convolution #67\n",
"no convolution #68\n",
"loading weights of convolution #69\n",
"loading weights of convolution #70\n",
"no convolution #71\n",
"loading weights of convolution #72\n",
"loading weights of convolution #73\n",
"no convolution #74\n",
"loading weights of convolution #75\n",
"loading weights of convolution #76\n",
"loading weights of convolution #77\n",
"loading weights of convolution #78\n",
"loading weights of convolution #79\n",
"loading weights of convolution #80\n",
"loading weights of convolution #81\n",
"no convolution #82\n",
"no convolution #83\n",
"loading weights of convolution #84\n",
"no convolution #85\n",
"no convolution #86\n",
"loading weights of convolution #87\n",
"loading weights of convolution #88\n",
"loading weights of convolution #89\n",
"loading weights of convolution #90\n",
"loading weights of convolution #91\n",
"loading weights of convolution #92\n",
"loading weights of convolution #93\n",
"no convolution #94\n",
"no convolution #95\n",
"loading weights of convolution #96\n",
"no convolution #97\n",
"no convolution #98\n",
"loading weights of convolution #99\n",
"loading weights of convolution #100\n",
"loading weights of convolution #101\n",
"loading weights of convolution #102\n",
"loading weights of convolution #103\n",
"loading weights of convolution #104\n",
"loading weights of convolution #105\n",
"WARNING:tensorflow:Compiled the loaded model, but the compiled metrics have yet to be built. `model.compile_metrics` will be empty until you train or evaluate the model.\n"
]
}
],
"source": [
"# create a YOLOv3 Keras model and save it to file\n",
"# based on https://github.com/experiencor/keras-yolo3\n",
"import struct\n",
3377
3378
3379
3380
3381
3382
3383
3384
3385
3386
3387
3388
3389
3390
3391
3392
3393
3394
3395
3396
3397
3398
3399
3400
3401
3402
3403
3404
3405
3406
3407
3408
3409
3410
3411
3412
3413
3414
3415
3416
3417
3418
3419
3420
3421
3422
3423
3424
3425
3426
3427
3428
3429
3430
3431
3432
3433
3434
3435
3436
3437
3438
3439
3440
3441
3442
3443
3444
3445
3446
3447
3448
3449
3450
3451
3452
3453
3454
3455
3456
3457
3458
3459
3460
3461
3462
3463
3464
3465
3466
3467
3468
3469
3470
3471
3472
3473
3474
3475
3476
3477
3478
3479
3480
3481
3482
3483
3484
3485
3486
3487
3488
3489
3490
3491
3492
3493
3494
3495
3496
3497
3498
3499
3500
3501
3502
3503
3504
3505
3506
3507
3508
3509
3510
3511
3512
3513
3514
3515
3516
3517
3518
3519
3520
3521
3522
3523
3524
3525
3526
3527
3528
3529
3530
3531
3532
3533
3534
3535
3536
3537
3538
3539
3540
"import numpy as np\n",
"from keras.layers import Conv2D\n",
"from keras.layers import Input\n",
"from keras.layers import BatchNormalization\n",
"from keras.layers import LeakyReLU\n",
"from keras.layers import ZeroPadding2D\n",
"from keras.layers import UpSampling2D\n",
"from keras.layers.merge import add, concatenate\n",
"from keras.models import Model\n",
"\n",
"def _conv_block(inp, convs, skip=True):\n",
" x = inp\n",
" count = 0\n",
" for conv in convs:\n",
" if count == (len(convs) - 2) and skip:\n",
" skip_connection = x\n",
" count += 1\n",
" if conv['stride'] > 1: x = ZeroPadding2D(((1,0),(1,0)))(x) # peculiar padding as darknet prefer left and top\n",
" x = Conv2D(conv['filter'],\n",
" conv['kernel'],\n",
" strides=conv['stride'],\n",
" padding='valid' if conv['stride'] > 1 else 'same', # peculiar padding as darknet prefer left and top\n",
" name='conv_' + str(conv['layer_idx']),\n",
" use_bias=False if conv['bnorm'] else True)(x)\n",
" if conv['bnorm']: x = BatchNormalization(epsilon=0.001, name='bnorm_' + str(conv['layer_idx']))(x)\n",
" if conv['leaky']: x = LeakyReLU(alpha=0.1, name='leaky_' + str(conv['layer_idx']))(x)\n",
" return add([skip_connection, x]) if skip else x\n",
"\n",
"def make_yolov3_model():\n",
" input_image = Input(shape=(None, None, 3))\n",
" # Layer 0 => 4\n",
" x = _conv_block(input_image, [{'filter': 32, 'kernel': 3, 'stride': 1, 'bnorm': True, 'leaky': True, 'layer_idx': 0},\n",
" {'filter': 64, 'kernel': 3, 'stride': 2, 'bnorm': True, 'leaky': True, 'layer_idx': 1},\n",
" {'filter': 32, 'kernel': 1, 'stride': 1, 'bnorm': True, 'leaky': True, 'layer_idx': 2},\n",
" {'filter': 64, 'kernel': 3, 'stride': 1, 'bnorm': True, 'leaky': True, 'layer_idx': 3}])\n",
" # Layer 5 => 8\n",
" x = _conv_block(x, [{'filter': 128, 'kernel': 3, 'stride': 2, 'bnorm': True, 'leaky': True, 'layer_idx': 5},\n",
" {'filter': 64, 'kernel': 1, 'stride': 1, 'bnorm': True, 'leaky': True, 'layer_idx': 6},\n",
" {'filter': 128, 'kernel': 3, 'stride': 1, 'bnorm': True, 'leaky': True, 'layer_idx': 7}])\n",
" # Layer 9 => 11\n",
" x = _conv_block(x, [{'filter': 64, 'kernel': 1, 'stride': 1, 'bnorm': True, 'leaky': True, 'layer_idx': 9},\n",
" {'filter': 128, 'kernel': 3, 'stride': 1, 'bnorm': True, 'leaky': True, 'layer_idx': 10}])\n",
" # Layer 12 => 15\n",
" x = _conv_block(x, [{'filter': 256, 'kernel': 3, 'stride': 2, 'bnorm': True, 'leaky': True, 'layer_idx': 12},\n",
" {'filter': 128, 'kernel': 1, 'stride': 1, 'bnorm': True, 'leaky': True, 'layer_idx': 13},\n",
" {'filter': 256, 'kernel': 3, 'stride': 1, 'bnorm': True, 'leaky': True, 'layer_idx': 14}])\n",
" # Layer 16 => 36\n",
" for i in range(7):\n",
" x = _conv_block(x, [{'filter': 128, 'kernel': 1, 'stride': 1, 'bnorm': True, 'leaky': True, 'layer_idx': 16+i*3},\n",
" {'filter': 256, 'kernel': 3, 'stride': 1, 'bnorm': True, 'leaky': True, 'layer_idx': 17+i*3}])\n",
" skip_36 = x\n",
" # Layer 37 => 40\n",
" x = _conv_block(x, [{'filter': 512, 'kernel': 3, 'stride': 2, 'bnorm': True, 'leaky': True, 'layer_idx': 37},\n",
" {'filter': 256, 'kernel': 1, 'stride': 1, 'bnorm': True, 'leaky': True, 'layer_idx': 38},\n",
" {'filter': 512, 'kernel': 3, 'stride': 1, 'bnorm': True, 'leaky': True, 'layer_idx': 39}])\n",
" # Layer 41 => 61\n",
" for i in range(7):\n",
" x = _conv_block(x, [{'filter': 256, 'kernel': 1, 'stride': 1, 'bnorm': True, 'leaky': True, 'layer_idx': 41+i*3},\n",
" {'filter': 512, 'kernel': 3, 'stride': 1, 'bnorm': True, 'leaky': True, 'layer_idx': 42+i*3}])\n",
" skip_61 = x\n",
" # Layer 62 => 65\n",
" x = _conv_block(x, [{'filter': 1024, 'kernel': 3, 'stride': 2, 'bnorm': True, 'leaky': True, 'layer_idx': 62},\n",
" {'filter': 512, 'kernel': 1, 'stride': 1, 'bnorm': True, 'leaky': True, 'layer_idx': 63},\n",
" {'filter': 1024, 'kernel': 3, 'stride': 1, 'bnorm': True, 'leaky': True, 'layer_idx': 64}])\n",
" # Layer 66 => 74\n",
" for i in range(3):\n",
" x = _conv_block(x, [{'filter': 512, 'kernel': 1, 'stride': 1, 'bnorm': True, 'leaky': True, 'layer_idx': 66+i*3},\n",
" {'filter': 1024, 'kernel': 3, 'stride': 1, 'bnorm': True, 'leaky': True, 'layer_idx': 67+i*3}])\n",
" # Layer 75 => 79\n",
" x = _conv_block(x, [{'filter': 512, 'kernel': 1, 'stride': 1, 'bnorm': True, 'leaky': True, 'layer_idx': 75},\n",
" {'filter': 1024, 'kernel': 3, 'stride': 1, 'bnorm': True, 'leaky': True, 'layer_idx': 76},\n",
" {'filter': 512, 'kernel': 1, 'stride': 1, 'bnorm': True, 'leaky': True, 'layer_idx': 77},\n",
" {'filter': 1024, 'kernel': 3, 'stride': 1, 'bnorm': True, 'leaky': True, 'layer_idx': 78},\n",
" {'filter': 512, 'kernel': 1, 'stride': 1, 'bnorm': True, 'leaky': True, 'layer_idx': 79}], skip=False)\n",
" # Layer 80 => 82\n",
" yolo_82 = _conv_block(x, [{'filter': 1024, 'kernel': 3, 'stride': 1, 'bnorm': True, 'leaky': True, 'layer_idx': 80},\n",
" {'filter': 255, 'kernel': 1, 'stride': 1, 'bnorm': False, 'leaky': False, 'layer_idx': 81}], skip=False)\n",
" # Layer 83 => 86\n",
" x = _conv_block(x, [{'filter': 256, 'kernel': 1, 'stride': 1, 'bnorm': True, 'leaky': True, 'layer_idx': 84}], skip=False)\n",
" x = UpSampling2D(2)(x)\n",
" x = concatenate([x, skip_61])\n",
" # Layer 87 => 91\n",
" x = _conv_block(x, [{'filter': 256, 'kernel': 1, 'stride': 1, 'bnorm': True, 'leaky': True, 'layer_idx': 87},\n",
" {'filter': 512, 'kernel': 3, 'stride': 1, 'bnorm': True, 'leaky': True, 'layer_idx': 88},\n",
" {'filter': 256, 'kernel': 1, 'stride': 1, 'bnorm': True, 'leaky': True, 'layer_idx': 89},\n",
" {'filter': 512, 'kernel': 3, 'stride': 1, 'bnorm': True, 'leaky': True, 'layer_idx': 90},\n",
" {'filter': 256, 'kernel': 1, 'stride': 1, 'bnorm': True, 'leaky': True, 'layer_idx': 91}], skip=False)\n",
" # Layer 92 => 94\n",
" yolo_94 = _conv_block(x, [{'filter': 512, 'kernel': 3, 'stride': 1, 'bnorm': True, 'leaky': True, 'layer_idx': 92},\n",
" {'filter': 255, 'kernel': 1, 'stride': 1, 'bnorm': False, 'leaky': False, 'layer_idx': 93}], skip=False)\n",
" # Layer 95 => 98\n",
" x = _conv_block(x, [{'filter': 128, 'kernel': 1, 'stride': 1, 'bnorm': True, 'leaky': True, 'layer_idx': 96}], skip=False)\n",
" x = UpSampling2D(2)(x)\n",
" x = concatenate([x, skip_36])\n",
" # Layer 99 => 106\n",
" yolo_106 = _conv_block(x, [{'filter': 128, 'kernel': 1, 'stride': 1, 'bnorm': True, 'leaky': True, 'layer_idx': 99},\n",
" {'filter': 256, 'kernel': 3, 'stride': 1, 'bnorm': True, 'leaky': True, 'layer_idx': 100},\n",
" {'filter': 128, 'kernel': 1, 'stride': 1, 'bnorm': True, 'leaky': True, 'layer_idx': 101},\n",
" {'filter': 256, 'kernel': 3, 'stride': 1, 'bnorm': True, 'leaky': True, 'layer_idx': 102},\n",
" {'filter': 128, 'kernel': 1, 'stride': 1, 'bnorm': True, 'leaky': True, 'layer_idx': 103},\n",
" {'filter': 256, 'kernel': 3, 'stride': 1, 'bnorm': True, 'leaky': True, 'layer_idx': 104},\n",
" {'filter': 255, 'kernel': 1, 'stride': 1, 'bnorm': False, 'leaky': False, 'layer_idx': 105}], skip=False)\n",
" model = Model(input_image, [yolo_82, yolo_94, yolo_106])\n",
" return model\n",
"\n",
"class WeightReader:\n",
" def __init__(self, weight_file):\n",
" with open(weight_file, 'rb') as w_f:\n",
" major,\t= struct.unpack('i', w_f.read(4))\n",
" minor,\t= struct.unpack('i', w_f.read(4))\n",
" revision, = struct.unpack('i', w_f.read(4))\n",
" if (major*10 + minor) >= 2 and major < 1000 and minor < 1000:\n",
" w_f.read(8)\n",
" else:\n",
" w_f.read(4)\n",
" transpose = (major > 1000) or (minor > 1000)\n",
" binary = w_f.read()\n",
" self.offset = 0\n",
" self.all_weights = np.frombuffer(binary, dtype='float32')\n",
"\n",
" def read_bytes(self, size):\n",
" self.offset = self.offset + size\n",
" return self.all_weights[self.offset-size:self.offset]\n",
"\n",
" def load_weights(self, model):\n",
" for i in range(106):\n",
" try:\n",
" conv_layer = model.get_layer('conv_' + str(i))\n",
" print(\"loading weights of convolution #\" + str(i))\n",
" if i not in [81, 93, 105]:\n",
" norm_layer = model.get_layer('bnorm_' + str(i))\n",
" size = np.prod(norm_layer.get_weights()[0].shape)\n",
" beta = self.read_bytes(size) # bias\n",
" gamma = self.read_bytes(size) # scale\n",
" mean = self.read_bytes(size) # mean\n",
" var = self.read_bytes(size) # variance\n",
" weights = norm_layer.set_weights([gamma, beta, mean, var])\n",
" if len(conv_layer.get_weights()) > 1:\n",
" bias = self.read_bytes(np.prod(conv_layer.get_weights()[1].shape))\n",
" kernel = self.read_bytes(np.prod(conv_layer.get_weights()[0].shape))\n",
" kernel = kernel.reshape(list(reversed(conv_layer.get_weights()[0].shape)))\n",
" kernel = kernel.transpose([2,3,1,0])\n",
" conv_layer.set_weights([kernel, bias])\n",
" else:\n",
" kernel = self.read_bytes(np.prod(conv_layer.get_weights()[0].shape))\n",
" kernel = kernel.reshape(list(reversed(conv_layer.get_weights()[0].shape)))\n",
" kernel = kernel.transpose([2,3,1,0])\n",
" conv_layer.set_weights([kernel])\n",
" except ValueError:\n",
" print(\"no convolution #\" + str(i))\n",
"\n",
" def reset(self):\n",
" self.offset = 0\n",
"\n",
"# define the model\n",
"model = make_yolov3_model()\n",
"# load the model weights\n",
"weight_reader = WeightReader('yolov3.weights')\n",
"# set the model weights into the model\n",
"weight_reader.load_weights(model)\n",
"# save the model to file\n",
"model.save('model.h5')"
]
},
3541
3542
3543
3544
3545
3546
3547
3548
3549
3550
3551
3552
3553
3554
3555
3556
3557
3558
3559
3560
3561
3562
3563
3564
3565
3566
3567
3568
3569
3570
3571
3572
3573
3574
3575
3576
3577
3578
3579
3580
3581
3582
3583
3584
3585
3586
3587
3588
3589
3590
3591
3592
3593
3594
3595
3596
3597
3598
3599
3600
3601
3602
3603
3604
3605
3606
3607
3608
3609
3610
3611
3612
3613
3614
3615
3616
3617
3618
3619
3620
3621
3622
3623
3624
3625
3626
3627
3628
3629
3630
3631
3632
3633
3634
3635
3636
3637
3638
3639
3640
3641
3642
3643
3644
3645
3646
3647
3648
3649
3650
3651
3652
3653
3654
3655
3656
3657
3658
3659
3660
3661
3662
3663
3664
3665
3666
3667
3668
3669
3670
3671
3672
3673
3674
3675
3676
3677
3678
3679
3680
3681
3682
3683
3684
3685
3686
3687
3688
3689
3690
3691
3692
3693
3694
3695
3696
3697
3698
3699
3700
3701
3702
3703
3704
3705
3706
3707
3708
3709
3710
3711
3712
3713
3714
3715
3716
3717
3718
3719
3720
3721
3722
3723
3724
3725
3726
3727
3728
3729
3730
3731
3732
3733
3734
3735
3736
3737
3738
3739
3740
3741
3742
3743
3744
3745
3746
3747
3748
3749
3750
3751
3752
3753
3754
3755
3756
3757
3758
3759
3760
3761
3762
3763
3764
3765
3766
3767
3768
3769
3770
3771
3772
3773
3774
3775
3776
3777
3778
3779
3780
3781
3782
3783
3784
3785
3786
3787
3788
3789
3790
3791
3792
3793
3794
3795
3796
3797
3798
3799
3800
3801
3802
3803
3804
3805
3806
3807
3808
3809
3810
3811
3812
3813
3814
3815
3816
3817
3818
3819
3820
3821
3822
3823
3824
3825
3826
3827
3828
3829
3830
3831
3832
3833
3834
3835
3836
3837
3838
3839
3840
3841
3842
3843
3844
3845
3846
3847
3848
3849
3850
3851
3852
3853
3854
3855
3856
3857
3858
3859
3860
3861
3862
3863
3864
3865
3866
3867
3868
3869
3870
3871
3872
3873
3874
3875
3876
3877
3878
3879
3880
3881
3882
3883
3884
3885
3886
3887
3888
3889
3890
3891
3892
3893
3894
3895
3896
3897
3898
3899
3900
3901
3902
3903
3904
3905
3906
3907
3908
3909
3910
3911
3912
3913
3914
3915
3916
3917
3918
3919
3920
3921
3922
3923
3924
3925
3926
3927
3928
3929
3930
3931
3932
3933
3934
3935
3936
3937
3938
3939
3940
3941
3942
3943
3944
3945
3946
3947
3948
3949
3950
3951
3952
3953
3954
3955
3956
3957
3958
3959
3960
3961
3962
3963
3964
3965
3966
3967
3968
3969
3970
3971
3972
3973
3974
3975
3976
3977
3978
3979
3980
3981
3982
3983
3984
3985
3986
3987
3988
3989
3990
3991
3992
3993
3994
3995
3996
3997
3998
3999
4000
{
"cell_type": "markdown",
"metadata": {},
"source": [
"Running the example may take a little less than one minute to execute on modern hardware.\n",
"\n",
"As the weight file is loaded, you will see debug information reported about what was loaded, output by the `WeightReader` class."
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"At the end of the run, the `model.h5` file is saved in your current working directory with approximately the same size as the original weight file (237MB), but ready to be loaded and used directly as a Keras model."
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"### Make a Prediction\n",
"\n",
"We need a new photo for object detection, ideally with objects that we know that the model knows about from the MSCOCO dataset.\n",
"\n",
"We will use a photograph of two elephants.\n"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"<img src='./Bilder/african-elephant.jpg'>"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"The first step is to load the Keras model. This might be the slowest part of making a prediction."
]
},
{
"cell_type": "raw",
"metadata": {},
"source": [
"# load yolov3 model\n",
"model = load_model('model.h5')"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"Next, we need to load our new photograph and prepare it as suitable input to the model. The model expects inputs to be color images with the square shape of $416\\times 416$ pixels.\n",
"\n",
"We can use the `load_img()` Keras function to load the image and the `target_size` argument to resize the image after loading. We can also use the `img_to_array()` function to convert the loaded `PIL` image object into a NumPy array, and then rescale the pixel values from $0-255$ to $0-1$ 32-bit floating point values.\n",
"\n"
]
},
{
"cell_type": "raw",
"metadata": {},
"source": [
"# load the image with the required size\n",
"image = load_img('./Bilder/african-elephant.jpg', target_size=(416, 416))\n",
"# convert to numpy array\n",
"image = img_to_array(image)\n",
"# scale pixel values to [0, 1]\n",
"image = image.astype('float32')\n",
"image /= 255.0"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"We will want to show the original photo again later, which means we will need to scale the bounding boxes of all detected objects from the square shape back to the original shape. As such, we can load the image and retrieve the original shape."
]
},
{
"cell_type": "raw",
"metadata": {},
"source": [
"# load the image to get its shape\n",
"image = load_img('./Bilder/african-elephant.jpg')\n",
"width, height = image.size"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"We can tie all of this together into a convenience function named `load_image_pixels()` that takes the filename and target size and returns the scaled pixel data ready to provide as input to the Keras model, as well as the original width and height of the image."
]
},
{
"cell_type": "code",
"execution_count": 17,
"metadata": {},
"outputs": [],
"source": [
"# load and prepare an image\n",
"def load_image_pixels(filename, shape):\n",
" # load the image to get its shape\n",
" image = load_img(filename)\n",
" width, height = image.size\n",
" # load the image with the required size\n",
" image = load_img(filename, target_size=shape)\n",
" # convert to numpy array\n",
" image = img_to_array(image)\n",
" # scale pixel values to [0, 1]\n",
" image = image.astype('float32')\n",
" image /= 255.0\n",
" # add a dimension so that we have one sample\n",
" image = expand_dims(image, 0)\n",
" return image, width, height"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"We can then call this function to load our photo of elephants."
]
},
{
"cell_type": "raw",
"metadata": {},
"source": [
"# define the expected input shape for the model\n",
"input_w, input_h = 416, 416\n",
"# define our new photo\n",
"photo_filename = 'african-elephant.jpg'\n",
"# load and prepare image\n",
"image, image_w, image_h = load_image_pixels(photo_filename, (input_w, input_h))"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"We can now feed the photo into the Keras model and make a prediction."
]
},
{
"cell_type": "raw",
"metadata": {},
"source": [
"# make prediction\n",
"yhat = model.predict(image)\n",
"# summarize the shape of the list of arrays\n",
"print([a.shape for a in yhat])"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"That’s it, at least for making a prediction. The complete example is listed below."
]
},
{
"cell_type": "raw",
"metadata": {},
"source": [
"# load yolov3 model and perform object detection\n",
"# based on https://github.com/experiencor/keras-yolo3\n",
"from numpy import expand_dims\n",
"from keras.models import load_model\n",
"from keras.preprocessing.image import load_img\n",
"from keras.preprocessing.image import img_to_array\n",
" \n",
"# load and prepare an image\n",
"def load_image_pixels(filename, shape):\n",
" # load the image to get its shape\n",
" image = load_img(filename)\n",
" width, height = image.size\n",
" # load the image with the required size\n",
" image = load_img(filename, target_size=shape)\n",
" # convert to numpy array\n",
" image = img_to_array(image)\n",
" # scale pixel values to [0, 1]\n",
" image = image.astype('float32')\n",
" image /= 255.0\n",
" # add a dimension so that we have one sample\n",
" image = expand_dims(image, 0)\n",
" return image, width, height\n",
" \n",
"# load yolov3 model\n",
"model = load_model('model.h5')\n",
"# define the expected input shape for the model\n",
"input_w, input_h = 416, 416\n",
"# define our new photo\n",
"photo_filename = 'zebra.jpg'\n",
"# load and prepare image\n",
"image, image_w, image_h = load_image_pixels(photo_filename, (input_w, input_h))\n",
"# make prediction\n",
"yhat = model.predict(image)\n",
"# summarize the shape of the list of arrays\n",
"print([a.shape for a in yhat])"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"Running the example returns a list of three NumPy arrays, the shape of which is displayed as output.\n",
"\n",
"These arrays predict both the bounding boxes and class labels but are encoded. They must be interpreted."
]
},
{
"cell_type": "raw",
"metadata": {},
"source": [
"[(1, 13, 13, 255), (1, 26, 26, 255), (1, 52, 52, 255)]"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"### Make a Prediction and Interpret Result\n",
"\n",
"The output of the model is, in fact, encoded candidate bounding boxes from three different grid sizes, and the boxes are defined the context of anchor boxes, carefully chosen based on an analysis of the size of objects in the MSCOCO dataset.\n",
"\n",
"The script provided by experiencor provides a function called `decode_netout()` that will take each one of the NumPy arrays, one at a time, and decode the candidate bounding boxes and class predictions. Further, any bounding boxes that don’t confidently describe an object (e.g. all class probabilities below a threshold) are ignored. We will use a probability of 60% or 0.6. The function returns a list of `BoundBox` instances that define the corners of each bounding box in the context of the input image shape and class probabilities."
]
},
{
"cell_type": "raw",
"metadata": {},
"source": [
"# define the anchors\n",
"anchors = [[116,90, 156,198, 373,326], [30,61, 62,45, 59,119], [10,13, 16,30, 33,23]]\n",
"# define the probability threshold for detected objects\n",
"class_threshold = 0.6\n",
"boxes = list()\n",
"for i in range(len(yhat)):\n",
" # decode the output of the network\n",
" boxes += decode_netout(yhat[i][0], anchors[i], class_threshold, input_h, input_w)"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"Next, the bounding boxes can be stretched back into the shape of the original image. This is helpful as it means that later we can plot the original image and draw the bounding boxes, hopefully detecting real objects.\n",
"\n",
"The experiencor script provides the `correct_yolo_boxes()` function to perform this translation of bounding box coordinates, taking the list of bounding boxes, the original shape of our loaded photograph, and the shape of the input to the network as arguments. The coordinates of the bounding boxes are updated directly."
]
},
{
"cell_type": "raw",
"metadata": {},
"source": [
"# correct the sizes of the bounding boxes for the shape of the image\n",
"correct_yolo_boxes(boxes, image_h, image_w, input_h, input_w)"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"The model has predicted a lot of candidate bounding boxes, and most of the boxes will be referring to the same objects. The list of bounding boxes can be filtered and those boxes that overlap and refer to the same object can be merged. We can define the amount of overlap as a configuration parameter, in this case, 50% or 0.5. This filtering of bounding box regions is generally referred to as non-maximal suppression and is a required post-processing step.\n",
"\n",
"The experiencor script provides this via the `do_nms()` function that takes the list of bounding boxes and a threshold parameter. Rather than purging the overlapping boxes, their predicted probability for their overlapping class is cleared. This allows the boxes to remain and be used if they also detect another object type."
]
},
{
"cell_type": "raw",
"metadata": {},
"source": [
"# suppress non-maximal boxes\n",
"do_nms(boxes, 0.5)"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"This will leave us with the same number of boxes, but only very few of interest. We can retrieve just those boxes that strongly predict the presence of an object: that is are more than 60% confident. This can be achieved by enumerating over all boxes and checking the class prediction values. We can then look up the corresponding class label for the box and add it to the list. Each box must be considered for each class label, just in case the same box strongly predicts more than one object.\n",
"\n",
"We can develop a `get_boxes()` function that does this and takes the list of boxes, known labels, and our classification threshold as arguments and returns parallel lists of boxes, labels, and scores."
]
},
{
"cell_type": "raw",
"metadata": {},
"source": [
"# get all of the results above a threshold\n",
"def get_boxes(boxes, labels, thresh):\n",
" v_boxes, v_labels, v_scores = list(), list(), list()\n",
" # enumerate all boxes\n",
" for box in boxes:\n",
" # enumerate all possible labels\n",
" for i in range(len(labels)):\n",
" # check if the threshold for this label is high enough\n",
" if box.classes[i] > thresh:\n",
" v_boxes.append(box)\n",
" v_labels.append(labels[i])\n",
" v_scores.append(box.classes[i]*100)\n",
" # don't break, many labels may trigger for one box\n",
" return v_boxes, v_labels, v_scores"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"We can call this function with our list of boxes.\n",
"\n",
"We also need a list of strings containing the class labels known to the model in the correct order used during training, specifically those class labels from the MSCOCO dataset. Thankfully, this is provided in the experiencor script."
]
},
{
"cell_type": "raw",
"metadata": {},
"source": [
"# define the labels\n",
"labels = [\"person\", \"bicycle\", \"car\", \"motorbike\", \"aeroplane\", \"bus\", \"train\", \"truck\",\n",
" \"boat\", \"traffic light\", \"fire hydrant\", \"stop sign\", \"parking meter\", \"bench\",\n",
" \"bird\", \"cat\", \"dog\", \"horse\", \"sheep\", \"cow\", \"elephant\", \"bear\", \"zebra\", \"giraffe\",\n",
" \"backpack\", \"umbrella\", \"handbag\", \"tie\", \"suitcase\", \"frisbee\", \"skis\", \"snowboard\",\n",
" \"sports ball\", \"kite\", \"baseball bat\", \"baseball glove\", \"skateboard\", \"surfboard\",\n",
" \"tennis racket\", \"bottle\", \"wine glass\", \"cup\", \"fork\", \"knife\", \"spoon\", \"bowl\", \"banana\",\n",
" \"apple\", \"sandwich\", \"orange\", \"broccoli\", \"carrot\", \"hot dog\", \"pizza\", \"donut\", \"cake\",\n",
" \"chair\", \"sofa\", \"pottedplant\", \"bed\", \"diningtable\", \"toilet\", \"tvmonitor\", \"laptop\", \"mouse\",\n",
" \"remote\", \"keyboard\", \"cell phone\", \"microwave\", \"oven\", \"toaster\", \"sink\", \"refrigerator\",\n",
" \"book\", \"clock\", \"vase\", \"scissors\", \"teddy bear\", \"hair drier\", \"toothbrush\"]\n",
"# get the details of the detected objects\n",
"v_boxes, v_labels, v_scores = get_boxes(boxes, labels, class_threshold)"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"Now that we have those few boxes of strongly predicted objects, we can summarize them."
]
},
{
"cell_type": "raw",
"metadata": {},
"source": [
"# summarize what we found\n",
"for i in range(len(v_boxes)):\n",
" print(v_labels[i], v_scores[i])"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"We can also plot our original photograph and draw the bounding box around each detected object. This can be achieved by retrieving the coordinates from each bounding box and creating a Rectangle object."
]
},
{
"cell_type": "raw",
"metadata": {},
"source": [
"box = v_boxes[i]\n",
"# get coordinates\n",
"y1, x1, y2, x2 = box.ymin, box.xmin, box.ymax, box.xmax\n",
"# calculate width and height of the box\n",
"width, height = x2 - x1, y2 - y1\n",
"# create the shape\n",
"rect = Rectangle((x1, y1), width, height, fill=False, color='white')\n",
"# draw the box\n",
"ax.add_patch(rect)"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"We can also draw a string with the class label and confidence."
]
},
{
"cell_type": "raw",
"metadata": {},
"source": [
"# draw text and score in top left corner\n",
"label = \"%s (%.3f)\" % (v_labels[i], v_scores[i])\n",
"pyplot.text(x1, y1, label, color='white')"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"The `draw_boxes()` function below implements this, taking the filename of the original photograph and the parallel lists of bounding boxes, labels and scores, and creates a plot showing all detected objects."
]
},
{
"cell_type": "code",
"execution_count": 20,
"metadata": {},
"outputs": [],
"source": [
"# draw all results\n",
"def draw_boxes(filename, v_boxes, v_labels, v_scores):\n",
" # load the image\n",
" data = pyplot.imread(filename)\n",
" # plot the image\n",
" pyplot.imshow(data)\n",
" # get the context for drawing boxes\n",
" ax = pyplot.gca()\n",
" # plot each box\n",
" for i in range(len(v_boxes)):\n",
" box = v_boxes[i]\n",
" # get coordinates\n",
" y1, x1, y2, x2 = box.ymin, box.xmin, box.ymax, box.xmax\n",
" # calculate width and height of the box\n",
" width, height = x2 - x1, y2 - y1\n",
" # create the shape\n",
" rect = Rectangle((x1, y1), width, height, fill=False, color='white')\n",
" # draw the box\n",
" ax.add_patch(rect)\n",
" # draw text and score in top left corner\n",
" label = \"%s (%.3f)\" % (v_labels[i], v_scores[i])\n",
" pyplot.text(x1, y1, label, color='white')\n",
" # show the plot\n",
" pyplot.show()"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"We can then call this function to plot our final result."
]
},
{
"cell_type": "raw",
"metadata": {},
"source": [
"# draw what we found\n",
"draw_boxes(photo_filename, v_boxes, v_labels, v_scores)"
]
},
{
"cell_type": "raw",
"metadata": {},
"source": [
"# draw what we found\n",
"draw_boxes(photo_filename, v_boxes, v_labels, v_scores)"
]
},
{
"cell_type": "code",
"execution_count": 21,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"WARNING:tensorflow:No training configuration found in the save file, so the model was *not* compiled. Compile it manually.\n",