bubbliiiing / yolov4-keras Goto Github PK

我想把训练好的模型转换到tf-serving上使用, pb格式的,请问有什么代码可以参考的吗? 谢谢~

显示在图上就是一个框对应多个标签。不是个例，是所有图片的测试结果都这样。请问是什么问题，怎么解决？感谢！

你好，请问一下我用了你的代码在titan v上训练并检测，FPS只有30几，和论文中的96FPS差的有点多，能请教下是什么原因吗

我想要训练自己的数据集，那么相关参数不需要改的嘛？

Hello, I want to know , which part is PAN?

老師您好~
想詢問input是規定32的倍數嗎?
如果網路上找到的dataset不符合，能改掉input的尺寸嗎?
如果不改，一樣使用416416但輸入影像為500375，這樣會影響結果嗎?

在 get_random_data_with_Mosaic 进行图像翻转操作时, 对应的box也进行了翻转, 而在 get_random_data 中, box没有进行对应的翻转, 这里是不是有问题, 还是我理解错了 ?

是不是代码有缺失

新手请教：
人脸识别的一些应用只需要传一张人脸照片 就能够识别了。
我们检测目标的时候为什么要传千张的图片？一张不行吗？

大大啥时候可以弄个yolov4 pytorch版本，上次听了您的bilibili上关于yolov3讲解，非常期待yolov4的版本，谢谢

训练了一个类别，大约200图片，把 载入预训练权重 那块代码注释掉，训练400epoch,loss为59，val_loss为48，但是测试框不出物体，是我哪里出错了吗

想請問如何測試整個資料夾的影像（.jpg或.mp4），並保存結果呢？
謝謝

UP您好，感谢您的教程和源码。请问是否可以将该项目改进一下，支持多GPU训练呢？

Your work is very valuable. Congratulations.

I cannot access the pan.baidu site. It does not allow access via Turkey. How can I access the files? Can you send a link to Google Drive.

Thank you in advance.

Best Regards.

您好！我想问一下，我是使用VOC数据集，利用yolov4_weight.h5这个权重文件，进行迁移学习，针对于VOC数据集使用k-means聚类得到的anchor box进行训练，我将余弦退火和标签平滑都设置为true，但是loss一直在50多高居不下。我随后又减小了学习率，使用前面的权重文件继续训练，但是loss仍然不见变小，请问您有什么意见嘛？哦，对了，我分别尝试了608和416的，都没有得到好的结果，但是使用您的VOC数据集的权重，就可以直接得到比较高的mAP。

大兄弟，哈哈。
1.我的数据命名都是从1自增的，用voc2yoyo4.py生成的train.txt中数字却不是自增的。这是不是无法和图片一一对应了。
2.运行voc_annotation.py时，显示UnicodeDecodeError: 'gbk' codec can't decode byte 0xae in position 96: illegal multibyte sequence；这个我查了都是从生成的文件中读数据，也没发现有非法字符。这是怎么回事呀

跟您請教一下，使用訓練好的Model，執行Video.py ，播放mp4檔案。
但FPS 落在15~17 Fps 。請問是否正常? 該如何提高? (GPU: GTX1060)
Thanks

yolov4-keras>python train.py
Using TensorFlow backend.
WARNING: Logging before flag parsing goes to stderr.
W0619 14:05:04.133064 51072 deprecation_wrapper.py:119] From C:\Users\assss\Anaconda3\envs\ID\lib\site-packages\keras\backend\tensorflow_backend.py:95: The name tf.reset_default_graph is d
eprecated. Please use tf.compat.v1.reset_default_graph instead.

W0619 14:05:04.134061 51072 deprecation_wrapper.py:119] From C:\Users\assss\Anaconda3\envs\ID\lib\site-packages\keras\backend\tensorflow_backend.py:98: The name tf.placeholder_with_default
is deprecated. Please use tf.compat.v1.placeholder_with_default instead.

W0619 14:05:04.144034 51072 deprecation_wrapper.py:119] From C:\Users\assss\Anaconda3\envs\ID\lib\site-packages\keras\backend\tensorflow_backend.py:102: The name tf.get_default_graph is de
precated. Please use tf.compat.v1.get_default_graph instead.

W0619 14:05:04.145031 51072 deprecation_wrapper.py:119] From C:\Users\assss\Anaconda3\envs\ID\lib\site-packages\keras\backend\tensorflow_backend.py:517: The name tf.placeholder is deprecat
ed. Please use tf.compat.v1.placeholder instead.

Create YOLOv4 model with 9 anchors and 5 classes.
W0619 14:05:04.146029 51072 deprecation_wrapper.py:119] From C:\Users\assss\Anaconda3\envs\ID\lib\site-packages\keras\backend\tensorflow_backend.py:4138: The name tf.random_uniform is depr
ecated. Please use tf.random.uniform instead.

2020-06-19 14:05:04.159481: I tensorflow/core/platform/cpu_feature_guard.cc:142] Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX2
2020-06-19 14:05:04.164326: I tensorflow/stream_executor/platform/default/dso_loader.cc:42] Successfully opened dynamic library nvcuda.dll
2020-06-19 14:05:04.265335: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1640] Found device 0 with properties:
name: GeForce RTX 2080 Ti major: 7 minor: 5 memoryClockRate(GHz): 1.635
pciBusID: 0000:01:00.0
2020-06-19 14:05:04.272995: I tensorflow/stream_executor/platform/default/dlopen_checker_stub.cc:25] GPU libraries are statically linked, skip dlopen check.
2020-06-19 14:05:04.276394: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1763] Adding visible gpu devices: 0
2020-06-19 14:05:04.834369: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1181] Device interconnect StreamExecutor with strength 1 edge matrix:
2020-06-19 14:05:04.837165: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1187] 0
2020-06-19 14:05:04.839025: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1200] 0: N
2020-06-19 14:05:04.841512: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1326] Created TensorFlow device (/job:localhost/replica:0/task:0/device:GPU:0 with 8694 MB memory) -> physica
l GPU (device: 0, name: GeForce RTX 2080 Ti, pci bus id: 0000:01:00.0, compute capability: 7.5)
W0619 14:05:05.187247 51072 deprecation_wrapper.py:119] From C:\Users\assss\Anaconda3\envs\ID\lib\site-packages\keras\backend\tensorflow_backend.py:1834: The name tf.nn.fused_batch_norm is
deprecated. Please use tf.compat.v1.nn.fused_batch_norm instead.

W0619 14:05:08.949218 51072 deprecation_wrapper.py:119] From C:\Users\assss\Anaconda3\envs\ID\lib\site-packages\keras\backend\tensorflow_backend.py:3976: The name tf.nn.max_pool is depreca
ted. Please use tf.nn.max_pool2d instead.

W0619 14:05:09.150654 51072 deprecation_wrapper.py:119] From C:\Users\assss\Anaconda3\envs\ID\lib\site-packages\keras\backend\tensorflow_backend.py:2018: The name tf.image.resize_nearest_n
eighbor is deprecated. Please use tf.compat.v1.image.resize_nearest_neighbor instead.

Load weights model_data/yolo4_weight.h5.
C:\Users\assss\Anaconda3\envs\ID\lib\site-packages\keras\engine\saving.py:1140: UserWarning: Skipping loading of weights for layer conv2d_110 due to mismatch in shape ((1, 1, 1024, 30) vs
(255, 1024, 1, 1)).
weight_values[i].shape))
C:\Users\assss\Anaconda3\envs\ID\lib\site-packages\keras\engine\saving.py:1140: UserWarning: Skipping loading of weights for layer conv2d_110 due to mismatch in shape ((30,) vs (255,)).
weight_values[i].shape))
C:\Users\assss\Anaconda3\envs\ID\lib\site-packages\keras\engine\saving.py:1140: UserWarning: Skipping loading of weights for layer conv2d_102 due to mismatch in shape ((1, 1, 512, 30) vs (
255, 512, 1, 1)).
weight_values[i].shape))
C:\Users\assss\Anaconda3\envs\ID\lib\site-packages\keras\engine\saving.py:1140: UserWarning: Skipping loading of weights for layer conv2d_102 due to mismatch in shape ((30,) vs (255,)).
weight_values[i].shape))
C:\Users\assss\Anaconda3\envs\ID\lib\site-packages\keras\engine\saving.py:1140: UserWarning: Skipping loading of weights for layer conv2d_94 due to mismatch in shape ((1, 1, 256, 30) vs (2
55, 256, 1, 1)).
weight_values[i].shape))
C:\Users\assss\Anaconda3\envs\ID\lib\site-packages\keras\engine\saving.py:1140: UserWarning: Skipping loading of weights for layer conv2d_94 due to mismatch in shape ((30,) vs (255,)).
weight_values[i].shape))
W0619 14:05:14.228083 51072 deprecation.py:323] From C:\Users\assss\Anaconda3\envs\ID\lib\site-packages\tensorflow\python\ops\array_ops.py:1354: add_dispatch_support..wrapper (from
tensorflow.python.ops.array_ops) is deprecated and will be removed in a future version.
Instructions for updating:
Use tf.where in 2.0, which has the same broadcast rule as np.where
Freeze the first 249 layers of total 370 layers.
W0619 14:05:15.692210 51072 deprecation_wrapper.py:119] From C:\Users\assss\Anaconda3\envs\ID\lib\site-packages\keras\optimizers.py:790: The name tf.train.Optimizer is deprecated. Please u
se tf.compat.v1.train.Optimizer instead.

Train on 6543 samples, val on 727 samples, with batch size 4.
W0619 14:05:22.052212 51072 deprecation_wrapper.py:119] From C:\Users\assss\Anaconda3\envs\ID\lib\site-packages\keras\callbacks.py:850: The name tf.summary.merge_all is deprecated. Please
use tf.compat.v1.summary.merge_all instead.

W0619 14:05:22.052212 51072 deprecation_wrapper.py:119] From C:\Users\assss\Anaconda3\envs\ID\lib\site-packages\keras\callbacks.py:853: The name tf.summary.FileWriter is deprecated. Please
use tf.compat.v1.summary.FileWriter instead.

Epoch 1/25
2020-06-19 14:05:27.713789: E tensorflow/core/grappler/optimizers/meta_optimizer.cc:502] shape_optimizer failed: Invalid argument: Subshape must have computed start >= end since stride is
negative, but is 0 and 2 (computed from start 0 and end 9223372036854775807 over shape with rank 2 and stride-1)
2020-06-19 14:05:28.038298: E tensorflow/core/grappler/optimizers/meta_optimizer.cc:502] remapper failed: Invalid argument: Subshape must have computed start >= end since stride is negativ
e, but is 0 and 2 (computed from start 0 and end 9223372036854775807 over shape with rank 2 and stride-1)
2020-06-19 14:05:29.572795: E tensorflow/core/grappler/optimizers/meta_optimizer.cc:502] layout failed: Invalid argument: Subshape must have computed start >= end since stride is negative,
but is 0 and 2 (computed from start 0 and end 9223372036854775807 over shape with rank 2 and stride-1)
2020-06-19 14:05:30.708524: E tensorflow/core/grappler/optimizers/meta_optimizer.cc:502] shape_optimizer failed: Invalid argument: Subshape must have computed start >= end since stride is
negative, but is 0 and 2 (computed from start 0 and end 9223372036854775807 over shape with rank 2 and stride-1)
2020-06-19 14:05:30.933520: E tensorflow/core/grappler/optimizers/meta_optimizer.cc:502] remapper failed: Invalid argument: Subshape must have computed start >= end since stride is negativ
e, but is 0 and 2 (computed from start 0 and end 9223372036854775807 over shape with rank 2 and stride-1)
2020-06-19 14:05:33.817409: E tensorflow/stream_executor/cuda/cuda_dnn.cc:329] Could not create cudnn handle: CUDNN_STATUS_ALLOC_FAILED
2020-06-19 14:05:33.820917: E tensorflow/stream_executor/cuda/cuda_dnn.cc:329] Could not create cudnn handle: CUDNN_STATUS_ALLOC_FAILED
Traceback (most recent call last):
File "train.py", line 254, in
callbacks=[logging, checkpoint, reduce_lr, early_stopping])
File "C:\Users\assss\Anaconda3\envs\ID\lib\site-packages\keras\legacy\interfaces.py", line 91, in wrapper
return func(*args, **kwargs)
File "C:\Users\assss\Anaconda3\envs\ID\lib\site-packages\keras\engine\training.py", line 1418, in fit_generator
initial_epoch=initial_epoch)
File "C:\Users\assss\Anaconda3\envs\ID\lib\site-packages\keras\engine\training_generator.py", line 217, in fit_generator
class_weight=class_weight)
File "C:\Users\assss\Anaconda3\envs\ID\lib\site-packages\keras\engine\training.py", line 1217, in train_on_batch
outputs = self.train_function(ins)
File "C:\Users\assss\Anaconda3\envs\ID\lib\site-packages\keras\backend\tensorflow_backend.py", line 2715, in call
return self._call(inputs)
File "C:\Users\assss\Anaconda3\envs\ID\lib\site-packages\keras\backend\tensorflow_backend.py", line 2675, in _call
fetched = self._callable_fn(*array_vals)
File "C:\Users\assss\Anaconda3\envs\ID\lib\site-packages\tensorflow\python\client\session.py", line 1458, in call
run_metadata_ptr)
tensorflow.python.framework.errors_impl.UnknownError: 2 root error(s) found.
(0) Unknown: Failed to get convolution algorithm. This is probably because cuDNN failed to initialize, so try looking to see if a warning log message was printed above.
[[{{node conv2d_1/convolution}}]]
[[loss/add_109/_4921]]
(1) Unknown: Failed to get convolution algorithm. This is probably because cuDNN failed to initialize, so try looking to see if a warning log message was printed above.
[[{{node conv2d_1/convolution}}]]
0 successful operations.
0 derived errors ignored.
我的GPU cuda版本是10.1 请问下您是否遇到过此类问题，谢谢

老师您好，请问如果想修改grid维度的话在哪里修改。

(sorry 按错 还没打完就发送了)
我试了一下你的 Yolov3 和 v4 但速度都是大概 FPS=6 的样子 (GTX1050)，难道v4不应该更快一点吗

学长你好

我想请问一下vision_for_anchors.py能发到GitHub上吗，就是绘制先验框的那个py文件，多谢！！

"ran out of memory trying to allocate 570.66MiB. The caller indicates that this is not a failure, but may mean that there could be performance gains if more memory were available."
请问这个问题应该如何解决

No module named 'utils.utils'怎么解决？

Is there any keras-yolov code for batch testing?

有什么办法可以减少模型的尺寸？

我用wider face数据集在您的v4上训练，分类也改成了只有人脸1类，训练50个epoch后做测试，运行get_gt_txt，得到的ground-truth中有很多图片中的人脸数量都超过20个，但运行get_dr_txt.py后，所有的预测结果中，在每张人脸图片标出来的人脸最多都不超过20个，这是什么原因，谢谢！

大神您好，跟您詢問，我將訓練好的Model 要，做PR Curve 要看效果好不好。
我依序執行get_dr_txt.py / get_gt_txt.py / 得到Input (detection-results/images-optional/ground-truth)
再執行get_map.py 想得到PR Curve, 但跑出來AP 為0。
比對Input 與Result 結果:
Input: detection-results 比ground-truth 多很多的Bounding box
detection-results 所框出的Bounding box 與我用predict.py 所跑出的結果，差異蠻多的。
想詢問是否是哪部分有做錯??
附件是相關的資料。
感謝。[

detection-results_convex_mirror(23).txt
ground-truth_convex_mirror(23).txt

老師您好~想詢問計算mAP得操作步驟?
有點搞不太懂

广告虽可耻，大家可以看看我复现的这版yolov4，精心设计了损失函数，训练了70000步就没出现过nan，mAP至少可达36%，欢迎点star：

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

你好，在按照您的步骤训练自己数据集的时候，修改epoch和freeze epoch后 loss一直nan 用您设定的值就正常。但是epoch 27就自动停止了，loss有65左右，单张图片检测想效果还行，用摄像头检测的话，误检严重。您有什么好的办法去修改优化吗

首先非常喜欢你的视频，声音很好听。
我在跑你的项目y4 和y3, 我没有什么修改，直接跑，同一个视频
y4识别出2个 目标 ，人 和椅子，y3则能多识别出三个目标 不知道为什么会导致这种差异
另外非常想知道 y4 fps 在我的电脑上只有0.6 y3 fps有0.8 我的电脑是不是该扔了，请教这两个项目最高fps大概能跑到多少，如果想优化应该调整那个部分能使它们跑的更快。非常感谢

这个工程的h5模型可否转为TensorRT模型。

大佬好，

我造了一些不同size的图片做训练集，在用yolo4训练之前还需要resize并处理好box的坐标吗？还是说无论什么图片都会在训练前被自动resize处理好？

叩谢

老师您好，我用自己的训练集（单class）训练完成后进行测试时发现每张图片预测的数量上限都在20，我找了一下yolo.py并没有发现有相关的设置，请问这个上限是在哪里设置的，谢谢！

如果想偵測影片(.mp4)中的特定範圍，該如何限制呢?

CUDA：10.0.130
系统：Ubuntu 16.04
tensorflow: 1.13
batchsize：8
问题：第一个eporch时loss就出现nan，采用COCO数据集，检测的目标物为：车辆、红绿灯、行人。

两块N卡训练，谢谢

大佬，请问怎么加DropBlock，‘DropBlock2D(block_size=2, keep_prob=0.8, name='DropBlock')’ 貌似不行

楼主你好，我把训练的.h5模型转换成saved_model格式。并且定义了输入输出，可是我部署到saved_model后预测报You must feed a value for placeholder tensor 'input_1' with dtype float and shape [?,?,?,3][[{{node input_1}}]] 麻烦说一下解决方法。我的输入和预测代码的输入是一致的。

大神，我样本的类别数和你不一样，所以我重新训练。如果从头开始训练，不用预训练模型model_path怎么改呢？

tensorflow.python.framework.errors_impl.InvalidArgumentError: slice index -1 of dimension 0 out of bounds.
for '{{node loss/yolo_loss_loss/strided_slice}} = StridedSlice[Index=DT_INT32, T=DT_INT32, begin_mask=0, ellipsis_mask=0, end_mask=0, new_axis_mask=0, shrink_axis_mask=1](loss/yolo_loss_loss/Shape, loss/yolo_loss_loss/strided_slice/stack, loss/yolo_loss_loss/strided_slice/stack_1, loss/yolo_loss_loss/strided_slice/stack_2)' with input shapes: [0], [1], [1], [1] and with computed input tensors: input[1] = <-1>, input[2] = <0>, input[3] = <1>.

您好，怎么样计算模型的flops值呢，希望您可以出一个代码。

如果我把mosaic设置为True，大概第一个epoch走了一半就nan了。
如果我把mosaic设置为False, 大概第三个epoch走了一大半后就nan了是什么问题呢.

請問如何將偵測完的影片檔(.mp4)，save成新的影片檔(.mp4)?
謝謝