A Keras implementation of YOLOv4 (Tensorflow backend)

Last update: Nov 29, 2022

Overview

keras-yolo4

请使用更完善的版本:

https://github.com/miemie2013/Keras-YOLOv4

Please visit here for more complete model:

https://github.com/miemie2013/Keras-YOLOv4

Introduction

A Keras implementation of YOLOv4 (Tensorflow backend) inspired by https://github.com/AlexeyAB/darknet.

Frame code from https://github.com/qqwweee/keras-yolo3.

Quick Start

Download YOLOv4 weights from yolov4.weights.
Convert the Darknet YOLOv4 model to a Keras model.
Run YOLOv4 detection.

python convert.py

Running convert.py will get keras yolov4 weight file yolo4_weight.h5.

Usage

Inference

python test.py

Todo

Debug train code with CIoU loss. Loss code from https://github.com/david8862/keras-YOLOv3-model-set.

Comments

Error while converting to keras model

I have trained yolo.weights (with my custom dataset having 3 classes) and the detections are coming fine while testing with that model. But when i am converting it to keras model, I am getting below error

File "convert.py", line 172, in <module>
    yolo4_model = Yolo4(score, iou, anchors_path, classes_path, model_path, weights_path)
  File "convert.py", line 157, in __init__
    self.load_yolo()
  File "convert.py", line 98, in load_yolo
    buffer=weights_file.read(weights_size * 4))
TypeError: buffer is too small for requested array

Could someone tell me why it is happening and what changes I should make

opened by vihari1729 11

any experiences about the training loss ?

thanks for the nice repo.

I am training yolov4 on coco dataset for only the first 7 classes with a split 70:30, any experiences or hints about the training loss when the model seems acceptable?

below is my traing log, seem a long time to train.

Train on 82801 samples, val on 35486 samples, with batch size 32.
Epoch 1/50

   1/2587 [..............................] - ETA: 28:56:58 - loss: 17721.8926
   2/2587 [..............................] - ETA: 16:33:45 - loss: 17243.4922
   3/2587 [..............................] - ETA: 12:32:28 - loss: 16760.1377
   4/2587 [..............................] - ETA: 10:10:04 - loss: 16299.9021
   5/2587 [..............................] - ETA: 9:24:35 - loss: 15857.3045 
   6/2587 [..............................] - ETA: 9:14:57 - loss: 15430.3268
   7/2587 [..............................] - ETA: 9:07:24 - loss: 15008.8902
   8/2587 [..............................] - ETA: 9:01:51 - loss: 14608.8591
   9/2587 [..............................] - ETA: 9:12:25 - loss: 14219.5904
  10/2587 [..............................] - ETA: 9:20:37 - loss: 13848.0055
  11/2587 [..............................] - ETA: 9:18:39 - loss: 13490.1777
  12/2587 [..............................] - ETA: 8:59:26 - loss: 13150.2271
  13/2587 [..............................] - ETA: 8:44:20 - loss: 12822.8066
  14/2587 [..............................] - ETA: 8:34:12 - loss: 12509.4747
  15/2587 [..............................] - ETA: 8:32:09 - loss: 12211.5387
  16/2587 [..............................] - ETA: 8:26:59 - loss: 11925.6479
  17/2587 [..............................] - ETA: 8:20:37 - loss: 11649.0334
  18/2587 [..............................] - ETA: 8:21:20 - loss: 11381.8880
  19/2587 [..............................] - ETA: 8:14:07 - loss: 11124.9040
  20/2587 [..............................] - ETA: 8:11:29 - loss: 10882.4443
  21/2587 [..............................] - ETA: 8:07:56 - loss: 10645.2164
  22/2587 [..............................] - ETA: 8:03:53 - loss: 10420.4502
  23/2587 [..............................] - ETA: 7:58:37 - loss: 10203.8660
  24/2587 [..............................] - ETA: 7:54:08 - loss: 9995.8066 
  25/2587 [..............................] - ETA: 7:50:39 - loss: 9794.8682
  26/2587 [..............................] - ETA: 7:48:39 - loss: 9599.6988
  27/2587 [..............................] - ETA: 7:45:55 - loss: 9411.8161
  28/2587 [..............................] - ETA: 7:42:36 - loss: 9234.4692
  29/2587 [..............................] - ETA: 7:39:51 - loss: 9060.8342
  30/2587 [..............................] - ETA: 7:35:15 - loss: 8894.1920
  31/2587 [..............................] - ETA: 7:33:45 - loss: 8733.4837
  32/2587 [..............................] - ETA: 7:31:51 - loss: 8577.9887
  33/2587 [..............................] - ETA: 7:29:49 - loss: 8425.8984
  34/2587 [..............................] - ETA: 7:27:43 - loss: 8280.7446
  35/2587 [..............................] - ETA: 7:26:30 - loss: 8139.6654
  36/2587 [..............................] - ETA: 7:26:01 - loss: 8004.0953
  37/2587 [..............................] - ETA: 7:26:13 - loss: 7872.3828
  38/2587 [..............................] - ETA: 7:24:52 - loss: 7746.5956
  39/2587 [..............................] - ETA: 7:23:28 - loss: 7624.6465
  40/2587 [..............................] - ETA: 7:29:19 - loss: 7506.7023
  41/2587 [..............................] - ETA: 7:35:16 - loss: 7391.7025
  42/2587 [..............................] - ETA: 7:34:34 - loss: 7280.5936
  43/2587 [..............................] - ETA: 7:31:58 - loss: 7172.5243
  44/2587 [..............................] - ETA: 7:29:56 - loss: 7067.7942
  45/2587 [..............................] - ETA: 7:30:26 - loss: 6965.9564
  46/2587 [..............................] - ETA: 7:31:08 - loss: 6867.1803
  47/2587 [..............................] - ETA: 7:34:20 - loss: 6771.7925

opened by Papageno2 5

"Unknown layer: Mish" while loading converted model from "yolo4_weight.h5"

Source code

from tensorflow import keras
new_model = keras.models.load_model('yolo4_weight.h5')

Error occurred:

Traceback (most recent call last):
  File "/home/source/keras-yolo4/mytest.py", line 13, in <module>
    new_model = keras.models.load_model(h5)
  File "/opt/conda/lib/python3.6/site-packages/tensorflow/python/keras/saving/save.py", line 146, in load_model
    return hdf5_format.load_model_from_hdf5(filepath, custom_objects, compile)
  File "/opt/conda/lib/python3.6/site-packages/tensorflow/python/keras/saving/hdf5_format.py", line 212, in load_model_from_hdf5
    custom_objects=custom_objects)
  File "/opt/conda/lib/python3.6/site-packages/tensorflow/python/keras/saving/model_config.py", line 55, in model_from_config
    return deserialize(config, custom_objects=custom_objects)
  File "/opt/conda/lib/python3.6/site-packages/tensorflow/python/keras/layers/serialization.py", line 89, in deserialize
    printable_module_name='layer')
  File "/opt/conda/lib/python3.6/site-packages/tensorflow/python/keras/utils/generic_utils.py", line 192, in deserialize_keras_object
    list(custom_objects.items())))
  File "/opt/conda/lib/python3.6/site-packages/tensorflow/python/keras/engine/network.py", line 1121, in from_config
    process_layer(layer_data)
  File "/opt/conda/lib/python3.6/site-packages/tensorflow/python/keras/engine/network.py", line 1105, in process_layer
    layer = deserialize_layer(layer_data, custom_objects=custom_objects)
  File "/opt/conda/lib/python3.6/site-packages/tensorflow/python/keras/layers/serialization.py", line 89, in deserialize
    printable_module_name='layer')
  File "/opt/conda/lib/python3.6/site-packages/tensorflow/python/keras/utils/generic_utils.py", line 181, in deserialize_keras_object
    config, module_objects, custom_objects, printable_module_name)
  File "/opt/conda/lib/python3.6/site-packages/tensorflow/python/keras/utils/generic_utils.py", line 166, in class_and_config_for_serialized_keras_object
    raise ValueError('Unknown ' + printable_module_name + ': ' + class_name)
ValueError: Unknown layer: Mish

opened by shangdibufashi 3

TypeError: buffer is too small for requested array

Loading weights. Weights Header: 1329865020 1348031555 1952981061 [1010723949] Converting 0 Traceback (most recent call last): File "convert.py", line 173, in yolo4_model = Yolo4(score, iou, anchors_path, classes_path, model_path, weights_path) File "convert.py", line 157, in init self.load_yolo() File "convert.py", line 130, in load_yolo buffer=weights_file.read(weights_size * 4)) TypeError: buffer is too small for requested array

opened by saanuregh 3

Your some layers are different from miemie2013's.

in @miemie2013's code

x = layers.ZeroPadding2D(padding=((1, 0), (1, 0)))(x)
x = conv2d_unit(x, i512, 3, strides=2)
s16 = conv2d_unit(x, i256, 1, strides=1)
x = conv2d_unit(x, i256, 1, strides=1)
x = stack_residual_block(x, i256, i256, n=8)
x = conv2d_unit(x, i256, 1, strides=1)
s16 = layers.Concatenate()([x, s16])
x = conv2d_unit(s16, i512, 1, strides=1)

where s16 is the output of Concatenate()

x = layers.UpSampling2D(2)(x)
s16 = conv2d_unit(s16, i256, 1, strides=1, act='leaky')
x = layers.Concatenate()([s16, x])

s16 then computer through conv2d_unit and concatenate with upsampling output x

as for your code:

y38 = DarknetConv2D_BN_Leaky(256, (1,1))(darknet.layers[204].output)
y38 = Concatenate()([y38, y19_upsample])

where darknet.layers[204].output is the output of DarknetConv2D_BN_Mish, not the concatenate layer darknet.layers[201].outputs.

So, I am curious which version is right, or both right?

opened by RuaHU 1

Lambda laye not compatible in tf2+

in tf2+, Lambda layer could not be used anymore, is there any solution to solve the issue especially th error:
ValueError: tf.function-decorated function tried to create variables on non-first call.

opened by robbebluecp 0
Batching with inference

I have converted a custom YOLO4 model to the tensorflow saved_model format. I can successfully use this for single image inference. Now when I feed in a batch of images for client side batching I get the results back in one list without separation of the results into frames so I can't work out which detections go with which frames. Can you provide any info on whether this behavior is expected or how I might remedy it?

opened by tylertroy 0

ValueError: Shapes (1, 1, 1024, 75) and (255, 1024, 1, 1) are incompatible

Traceback (most recent call last):
  File "test.py", line 58, in <module>
    yolo4_model.load_weights(model_path)
  File "/usr/local/lib/python3.6/dist-packages/keras/engine/saving.py", line 492, in load_wrapper
    return load_function(*args, **kwargs)
  File "/usr/local/lib/python3.6/dist-packages/keras/engine/network.py", line 1230, in load_weights
    f, self.layers, reshape=reshape)
  File "/usr/local/lib/python3.6/dist-packages/keras/engine/saving.py", line 1237, in load_weights_from_hdf5_group
    K.batch_set_value(weight_value_tuples)
  File "/usr/local/lib/python3.6/dist-packages/keras/backend/tensorflow_backend.py", line 2960, in batch_set_value
    tf_keras_backend.batch_set_value(tuples)
  File "/home/mayur/.local/lib/python3.6/site-packages/tensorflow_core/python/keras/backend.py", line 3323, in batch_set_value
    x.assign(np.asarray(value, dtype=dtype(x)))
  File "/home/mayur/.local/lib/python3.6/site-packages/tensorflow_core/python/ops/resource_variable_ops.py", line 819, in assign
    self._shape.assert_is_compatible_with(value_tensor.shape)
  File "/home/mayur/.local/lib/python3.6/site-packages/tensorflow_core/python/framework/tensor_shape.py", line 1110, in assert_is_compatible_with
    raise ValueError("Shapes %s and %s are incompatible" % (self, other))
ValueError: Shapes (1, 1, 1024, 75) and (255, 1024, 1, 1) are incompatible

opened by mayurmahurkar 5

[BUG] iou computation in preprocess_true_boxes() should not use scaled bbox

in model.py, preprocess_true_boxes(): for i in range(3): anchors_xywh = np.zeros((3, 4)) anchors_xywh[:, 0:2] = np.floor(bbox_xywh_scaled[i, 0:2]).astype(np.int32) + 0.5 anchors_xywh[:, 2:4] = anchors[i] iou_scale = bbox_iou_data(bbox_xywh_scaled[i][np.newaxis, :], anchors_xywh) iou.append(iou_scale)

since anchors_xywh is in original scale, we shouldn't use bbox_xywh instead of bbox_xywh_scaled to compute iou?

opened by maoqingy 0
What do the four parameters after loss mean during training, loss of each category or other else?

Epoch 89/150 loss: [70.428894][6.8589654][49.6575623][13.912364] 1234/1234 [==============================] - 638s 517ms/step - loss: 48.12988.202549][5.05134583][33.3962708][9.7549324]]]]]]] Epoch end eval mAP on weight logs/222/ep089-loss48.130.h5 Epoch 89 mAP 0.2658832599269524 Epoch 90/150 loss: [59.5338707][5.99079227][46.7998619][6.74321651] 1234/1234 [==============================] - 655s 531ms/step - loss: 47.88505.1289482][3.23023152][27.289856][4.60885954]]]]] Epoch end eval mAP on weight logs/222/ep090-loss47.885.h5 Epoch 90 mAP 0.26986168503362695

opened by Kuuuo 0

Owner

GitHub

YOLOv4 / Scaled-YOLOv4 / YOLO - Neural Networks for Object Detection (Windows and Linux version of Darknet )

Yolo v4, v3 and v2 for Windows and Linux (neural networks for object detection) Paper YOLO v4: https://arxiv.org/abs/2004.10934 Paper Scaled YOLO v4:

20.2k Jan 9, 2023

A repository that shares tuning results of trained models generated by TensorFlow / Keras. Post-training quantization (Weight Quantization, Integer Quantization, Full Integer Quantization, Float16 Quantization), Quantization-aware training. TensorFlow Lite. OpenVINO. CoreML. TensorFlow.js. TF-TRT. MediaPipe. ONNX. [.tflite,.h5,.pb,saved_model,tfjs,tftrt,mlmodel,.xml/.bin, .onnx]

PINTO_model_zoo Please read the contents of the LICENSE file located directly under each folder before using the model. My model conversion scripts ar

2.4k Jan 5, 2023

Cascaded Pyramid Network (CPN) based on Keras (Tensorflow backend)

ML2 Takehome Project Reimplementing the paper: Cascaded Pyramid Network for Multi-Person Pose Estimation Dataset The model uses the COCO dataset which

1 Nov 22, 2021

Object tracking implemented with YOLOv4, DeepSort, and TensorFlow.

Object tracking implemented with YOLOv4, DeepSort, and TensorFlow. YOLOv4 is a state of the art algorithm that uses deep convolutional neural networks to perform object detections. We can take the output of YOLOv4 feed these object detections into Deep SORT (Simple Online and Realtime Tracking with a Deep Association Metric) in order to create a highly accurate object tracker.

1.1k Dec 29, 2022

PyTorch ,ONNX and TensorRT implementation of YOLOv4

4.2k Jan 1, 2023

Implementation of "Scaled-YOLOv4: Scaling Cross Stage Partial Network" using PyTorch framwork.

YOLOv4-large This is the implementation of "Scaled-YOLOv4: Scaling Cross Stage Partial Network" using PyTorch framwork. YOLOv4-CSP YOLOv4-tiny YOLOv4-

2k Jan 2, 2023

This is an implementation of Googles Yogi-Optimizer in Keras (tf.keras)

Yogi-Optimizer_Keras This is an implementation of Googles Yogi-Optimizer in Keras (tf.keras) The NeurIPS-Paper can be found here: http://papers.nips.c

14 Sep 13, 2022

Keras udrl - Keras implementation of Upside Down Reinforcement Learning

keras_udrl Keras implementation of Upside Down Reinforcement Learning This is me

7 Jan 24, 2022

这是一个mobilenet-yolov4-lite的库，把yolov4主干网络修改成了mobilenet，修改了Panet的卷积组成，使参数量大幅度缩小。

YOLOV4：You Only Look Once目标检测模型-修改mobilenet系列主干网络-在Keras当中的实现 2021年2月8日更新：加入letterbox_image的选项，关闭letterbox_image后网络的map一般可以得到提升。

65 Dec 1, 2022

YOLTv4 builds upon YOLT and SIMRDWN, and updates these frameworks to use the most performant version of YOLO, YOLOv4

YOLTv4 builds upon YOLT and SIMRDWN, and updates these frameworks to use the most performant version of YOLO, YOLOv4. YOLTv4 is designed to detect objects in aerial or satellite imagery in arbitrarily large images that far exceed the ~600×600 pixel size typically ingested by deep learning object detection frameworks.

161 Jan 6, 2023

I tried to apply the CAM algorithm to YOLOv4 and it worked.

YOLOV4：You Only Look Once目标检测模型在pytorch当中的实现 2021年2月7日更新：加入letterbox_image的选项，关闭letterbox_image后网络的map得到大幅度提升。目录性能情况 Performance 实现的内容 Achievement

55 Dec 5, 2022

People movement type classifier with YOLOv4 detection and SORT tracking.

Movement classification The goal of this project would be movement classification of people, in other words, walking (normal and fast) and running. Yo

4 Sep 21, 2021

Vehicles Counting using YOLOv4 + DeepSORT + Flask + Ngrok

A project for counting vehicles using YOLOv4 + DeepSORT + Flask + Ngrok

37 Dec 16, 2022

Implementing yolov4 target detection and tracking based on nao robot

6 Apr 19, 2022

A tutorial on training a DarkNet YOLOv4 model for the CrowdHuman dataset

YOLOv4 CrowdHuman Tutorial This is a tutorial demonstrating how to train a YOLOv4 people detector using Darknet and the CrowdHuman dataset. Table of c

118 Nov 10, 2022

YOLOv4-v3 Training Automation API for Linux

This repository allows you to get started with training a state-of-the-art Deep Learning model with little to no configuration needed! You provide your labeled dataset or label your dataset using our BMW-LabelTool-Lite and you can start the training right away and monitor it in many different ways like TensorBoard or a custom REST API and GUI. NoCode training with YOLOv4 and YOLOV3 has never been so easy.

626 Dec 31, 2022