yihanglou / fast-rcnn-train-another-dataset Goto Github PK

Hello,

How can we pick-up a pre-trained model and re-train it on a custom data-set (user defined data-set)? I have selected the PVANET+ (Compressed) because as I see it shows a good combination of speed and mAP.

Questions like:

• The Train/Test images should be annoated to what format?
• How to created the LMDB files for PVANet after annotation? By standard Caffe tools or you have your own tool?
• How to modify CFG / Prototxt files?

Considering that I gonna fine-tune on existing trained .caffemodels, so I must get a high accuracy even on a small dataset. Yes?

出错的是train.py里的snapshot函数：
net.params['bbox_pred'][0].data[...] =
(net.params['bbox_pred'][0].data *
self.bbox_stds[:, np.newaxis])
我调试了下，net.params['bbox_pred'][0].data 它的shape是 8 * 32 * 1 * 1， self.bbox_stds[:, np.newaxis 它的shape是 8 * 1， 它俩相乘后的结果，shape是： 8 * 32 * 8 * 1 ，把这个结果保存到 net.params['bbox_pred'][0].data[...] 时会出错，因为net.params['bbox_pred'][0].data[...]的shape是 8 * 32 * 1 * 1，不知道你遇到过这个问题没？

Running with the raw code and here is the following bug:

Traceback (most recent call last):
File "./tools/train_net.py", line 15, in
from datasets.factory import get_imdb
File "/home/liujs/pycaffe/py-faster-rcnn-master/tools/../lib/datasets/init.py", line 7, in
from .kakou import kakou
File "/home/liujs/pycaffe/py-faster-rcnn-master/tools/../lib/datasets/kakou.py", line 21, in
class kakou(datasets.imdb):
TypeError: Error when calling the metaclass bases
module.init() takes at most 2 arguments (3 given)

Would you please share the opinion to fix it?

您好！按照您的博客fast rcnn训练自己的数据集，我遇到了列表越界的问题。由于我的GT数据是写在TXT文件当中，只有背景和目标两类。但是一张图片中包含了几个目标，所以我的GT文件格式是这样的：xxx\0001.jpg 2 12 34 35 89 23 34 56 132 。于是出现了错误：y2=float(split_line[5+i*4]) IndexError:list index out of range。请问您的训练样本都是一张图一个目标的么？

如果，一张图片中有多辆车，在ImageList_Version_S_AddData.txt里面，每辆车的坐标是怎么样子的？
是下面这样吗？

ImageVersion/2015.jpg 2 118 46 208 395 18 6 200 300

（其中，118 46 208 395是第1辆车的坐标，18 6 200 300是第2辆车的坐标）

在博客第二篇中，你说到用EdgeBox的办法得到EdgeBox_Version_S_AddData.mat文件，不知道具体是怎么做到的，可否在github上给出相应的文件或者更新博客说明一下怎么生成的EdgeBox_Version_S_AddData.mat这个文件，非常感谢，此外对您这个项目的文件结构也不是很清晰，能够给出最后的文件结构吗？再次感谢

In my train data file, a txt file contains all path of input images and the boundary of it. It likes "1.png 1 1 2 3 4". This image path at function read not only path but also boundary information. So it is an error says path does not exists. How could I fix it?