Faster-ILOD、maskrcnn_ Benchmark trains its own VOC data set and problem summary

One 、 Train yourself to mark VOC Data sets

Use Faster-ILOD Train yourself to annotate data sets , The dataset has been converted to VOC Format
I have been able to train smoothly before VOC2007 Data sets , May refer to
Faster-ILOD、maskrcnn_benchmark Installation process and problems encountered
Mainly modify the following documents ：

1.maskrcnn_benchmark/config/paths_catalog.py

The corresponding path is modified to the path where your dataset is located

        "voc_2007_train": {
    
            "data_dir": "/data/taxi_data/VOC2007_all",
            "split": "train"
        },
        "voc_2007_val": {
    
            "data_dir": "/data/taxi_data/VOC2007_all",
            "split": "val"
        },
        "voc_2007_test": {
    
            "data_dir": "/data/taxi_data/VOC2007_all",
            "split": "test"
        },

2.maskrcnn_benchmark/data/voc.py

Modify to include categories in the dataset I mainly identify taxis and private cars

    # CLASSES = ("__background__ ", "aeroplane", "bicycle", "bird", "boat", "bottle", "bus", "car", "cat", "chair", "cow",
    # "diningtable", "dog", "horse", "motorbike", "person", "pottedplant", "sheep", "sofa", "train", "tvmonitor")
    CLASSES = ("__background__ ", "Taxi","Other Car")

3. modify configs/e2e_faster_rcnn_R_50_C4_1x.yaml

Modify the number of categories , I trained both classes at once .

NUM_CLASSES: 3 # total classes 
NAME_OLD_CLASSES: []
#NAME_NEW_CLASSES: ["aeroplane", "bicycle", "bird", "boat", "bottle", "bus", "car", "cat", "chair", "cow", "diningtable", "dog",
# "horse", "motorbike", "person","pottedplant", "sheep", "sofa", "train","tvmonitor"]
# NAME_EXCLUDED_CLASSES: [ ]
NAME_NEW_CLASSES: ["Taxi","Other Car"]
NAME_EXCLUDED_CLASSES: []

4. function

python tools/train_first_step.py --config-file="./configs/e2e_faster_rcnn_R_50_C4_1x.yaml"

Two 、 Have a problem

1. There is no counterpart Taxi_train.txt Wait for the documents

Because it is incremental learning , It will be processed for each category , Originally downloaded voc The file contains the corresponding train、test etc. txt file , But the data set of self annotation is converted to voc Format data set is converted to train.txt,test.txt,trainval.txt as well as val.txt. There is no counterpart txt file , Therefore, the corresponding file should be generated .

Observe the corresponding txt file , Each row corresponds to two columns , One column is the name of the picture , The other column is a single digit , Expected value is 1.-1.0.
-1 It means that this class does not exist in this picture
1 Bits exist in this picture
0 For difficult samples
（ Personally, I think so , Wrong, welcome to correct ）

according to xml Document and train.txt,test.txt,val.txt Convert the corresponding category . The code is as follows ：
When I generate the corresponding file, I only write the name of the image with this category , When reading data, the code is changed if an error is reported , You can modify the code , Consistent with the original data set .

names = locals()
dir_path = "D:/achenf/data/jiaotong_data/1taxi_train_data/VOC/VOC2007/"
#  Read train.txt/test.txt/val.txt file 
f = open("D:/achenf/data/jiaotong_data/1taxi_train_data/VOC/VOC2007/ImageSets/Main/train.txt", 'r')
#  Include categories 
classes = ['Taxi','Other Car']
#  Write document name 
txt_path = dir_path+"by_classes/"+"Other Car_train.txt"
fp_w = open(txt_path, 'w')
for line in f.readlines():
    if not line:
        break
    n= line[:-1]
    xmlpath=dir_path+"Annotations/"+n+'.xml'
    fp = open(xmlpath)

    xmllines = fp.readlines()
    ind_start = []
    ind_end = []
    lines_id_start = xmllines[:]
    lines_id_end = xmllines[:]
    #  Modify the corresponding category Other Car/Taxi
    classes1 = ' <name>Other Car</name>\n'
    while " <object>\n" in lines_id_start:
        a = lines_id_start.index(" <object>\n")
        ind_start.append(a)
        lines_id_start[a] = "delete"

    while " </object>\n" in lines_id_end:
        b = lines_id_end.index(" </object>\n")
        ind_end.append(b)
        lines_id_end[b] = "delete"

    # names Store all the object block 
    i = 0
    for k in range(0, len(ind_start)):
        names['block%d' % k] = []
        for j in range(0, len(classes)):
            if classes[j] in xmllines[ind_start[i] + 1]:
                a = ind_start[i]
                for o in range(ind_end[i] - ind_start[i] + 1):
                    names['block%d' % k].append(xmllines[a + o])
                break
        i += 1
    #  Search for... In a given class , If there is, then , write in object Piece of information 
    a = 0
    if len(ind_start)>0:
        # xml head 
        string_start = xmllines[0:ind_start[0]]
        # xml tail 
        string_end = [xmllines[len(xmllines) - 1]]
        flag = False
        for k in range(0, len(ind_start)):
            if classes1 in names['block%d' % k]:
                flag = True
                a += 1
                string_start += names['block%d' % k]
        string_start += string_end
        #  If there is a write 
        if flag:
            fp_w.write(n+" "+"1"+'\n')
    fp.close()
fp_w.close()

Get the following documents ：

2.x[2] == ‘0’ Subscript exceeded

When dividing each line , If the value is 1, Source dataset txt The document is decomposed into [‘000000’,‘’,‘1’], The file I generated is decomposed into [‘000000’,‘1’], Because there are no difficult samples , Comment these lines directly

    for i in range(len(buff)):
        x = buff[i]
        x = x.split(' ')
        if x[1] == '-1':
            pass
        # elif x[2] == '0': # include difficult level object
        # if self.is_train:
        # pass
        # else:
        # img_ids_per_category.append(x[0])
        # self.ids.append(x[0])
        else:
            img_ids_per_category.append(x[0])
            self.ids.append(x[0])

And my xml Nor does it exist in the file difficult The label of , So this line of code difficult = int(obj.find("difficult").text) == 1 You will also report mistakes. , So directly difficult The assignment is False

for obj in target.iter("object"):

     difficult = False
     # difficult = int(obj.find("difficult").text) == 1
     if not self.keep_difficult and difficult:
         continue

3.ValueError: invalid literal for int() with base 10: ‘0.0’

Report the mistake on this trip bndbox = tuple(map(lambda x: x - TO_REMOVE, list(map(int, box)))), Should be str Go straight to int The problem of , So we first convert it into float, In the transformation of int, Revised as follows ：bndbox = tuple(map(lambda x: x - TO_REMOVE, list(map(int, map(float,box)))))
Just ok 了
Learn a new knowledge point ,map（） Function can realize batch data type conversion .

May refer to ： Python3 of use map() Convert data types in batches , for example str turn float

4.Pytorch Report errors CUDA error: device-side assert triggered

Reference resources ：https://blog.csdn.net/veritasalice/article/details/111917185