Issue calthec data set training

[Marouane1993]

I have changed the calthec data set format to pascal_voc, but i got the following error:

Traceback (most recent call last): File "./tools/trainval_net.py", line 105, in imdb, roidb = combined_roidb(args.imdb_name) File "./tools/trainval_net.py", line 76, in combined_roidb roidbs = [get_roidb(s) for s in imdb_names.split('+')] File "./tools/trainval_net.py", line 73, in get_roidb roidb = get_training_roidb(imdb) File "/raid/users/mhamda/tf-faster-rcnn-master/tools/../lib/model/train_val.py", line 328, in get_training_roidb imdb.append_flipped_images() File "/raid/users/mhamda/tf-faster-rcnn-master/tools/../lib/datasets/imdb.py", line 118, in append_flipped_images assert (boxes[:, 2] >= boxes[:, 0]).all() AssertionError

[rnsandeep]

Some of your bounding boxes could have been wrong.

Use this script to filter the wrong annotations.

!/usr/bin/env python

import os from xml.dom import minidom os.environ['GLOG_minloglevel'] = '2' import os, sys, cv2 import argparse import shutil import sys ,os from os import listdir from os.path import isfile, join import multiprocessing from time import sleep

CLASSES = ('background','A','B','C') def read_xml(xml_path, new_path): flag = True doc = minidom.parse(xml_path) objects = doc.getElementsByTagName("object") fname = doc.getElementsByTagName("filename")[0].firstChild.data size = doc.getElementsByTagName("size")[0] width = size.getElementsByTagName("width")[0].firstChild.data height = size.getElementsByTagName("height")[0].firstChild.data labelledObjects = dict() c = 0 for obj in objects: if obj.getElementsByTagName("name")[0].firstChild.data in CLASSES: c = c+1 if not c>0: flag = False return flag

for obj in objects:
    name = obj.getElementsByTagName("name")[0].firstChild.data
    xmin = obj.getElementsByTagName("xmin")[0].firstChild.data
    ymin = obj.getElementsByTagName("ymin")[0].firstChild.data
    xmax = obj.getElementsByTagName("xmax")[0].firstChild.data
    ymax = obj.getElementsByTagName("ymax")[0].firstChild.data

    if int(ymin) >= int(height):
        continue

    if int(ymax) >= int(height):
        ymax = height

    if int(xmin) >= int(xmax):
        continue

    if  int(ymin) >= int(ymax):
        continue

    if  int(xmin) >= int(width):
        continue

    if int(xmax) >= int(width) :
        xmax = width

    if int(xmin) <=0: 
        xmin = str(1)
    if int(ymin) <=0:
        ymin = str(1)
    if int(xmax) <=0: 
        continue
    if int(ymax) <=0:
        continue

    box = [xmin, ymin, xmax, ymax]
    bb = []
    for b in box:
        if int(b)  ==0:
            b = 1
        bb.append(b)
    box = (bb[0], bb[1], bb[2] , bb[3])
    if name not in labelledObjects:
        labelledObjects[name] = []
        labelledObjects[name].append(box)
    else:
        labelledObjects[name].append(box)

newxml = write_xml(fname, labelledObjects, width, height)
g = open(new_path,'w')
g.write(newxml)
g.close()
return flag

def write_xml(file_name, objects, width, height): ann = "PASCAL"+file_name+"0PASCAL DBPASCAL DB Celeb IMDBPASCALPASCALPASCAL" ann += ""+width+""+ height+"3" for obj in objects: for ins in objects[obj]: ann += "" ann += "" return ann

if name == 'main': mypath = sys.argv[1] outputpath = sys.argv[2] os.makedirs(outputpath) images = [mypath+'/'+f for f in listdir(mypath) if isfile(join(mypath, f))] count = 0
totalCount = 0 jobs = [] for image in images: p = multiprocessing.Process(target=read_xml, args=(image, outputpath+'/'+image.split('/')[-1], )) jobs.append(p) totalCount = totalCount + 1 p.start()

[minooisbusy]

hello @rnsandeep First of all, thank you very much for writing this solution. There is one thing I would like to ask. Can I rewrite this code to the correct tab size? I can not understand this code because of the tab size. Thank you.

[rnsandeep]

sure please do that.