def test_e_fp(filename,expected_count=10): init_gamera() c = Classifier_with_remove() c.set_k(1) c.change_features(["volume64regions"]) ci = c.classify_image(filename) #files = ["mergedyn2.xml", "mergedyn.xml","only-dynamics.xml", # "newtrain-dynamic.xml", "preomr.xml"] files = ["preomr.xml","preomr_edited.xml","preomr_edited_cnn.xml"] import os.path # try to match with different trainingsets. for dynamic in ([ d for d in files if os.path.isfile(d) ]): ci.load_new_training_data(dynamic) print "%s - count_of_training=%d, k=%d"%(dynamic,len(c.stats),c.k) result = {} # Push into buckets based on the count of found glyphs. csv = {} sys.stdout.flush() # Try with different epsilon for false_positives: e_fp for e_fp in arange(0.01,1.01,0.01): c.e_fp=e_fp count = len(ci.classified_glyphs()) # Init bucket. if not result.has_key(count): result[count] = [] result[count].append((e_fp,c.d_t())) csv[e_fp] = count # Find the best match to the wanted result. k,res,diff = find_nearest(result,expected_count) confid = [ (len(v),key,v[0][0],v[0][1]) for key,v in result.iteritems() ] confid2 = [ (key,len(v)) for key,v in result.iteritems() ] confid.sort(reverse=True) confid2.sort() print "efp,count" for e_fp,c in sorted(csv.iteritems()): print "%s,%s"%(e_fp,c) print print "count,spansize" for count,spansize in confid2: print "%s,%s"%(count,spansize) return ret = [] for i in range(0,min(10,len(confid))+1): ret.append(confid[i]) if not result.has_key(expected_count): print "Never found the desired amount with %s"%dynamic print "Found in %d(%d): %s"%(k,diff,[r for r in res]) rgbimg = ci.image.to_rgb() cg = ci.classified_glyphs(res[0].d_t) [outline(rgbimg,g,3.0,RGBPixel(255,0,0)) for g in cg] rgbimg.save_PNG("class_%s_%s.png"%(filename,dynamic)) print
def test_e_fp(filename, expected_count=10): init_gamera() c = Classifier_with_remove() c.set_k(1) c.change_features(["volume64regions"]) ci = c.classify_image(filename) #files = ["mergedyn2.xml", "mergedyn.xml","only-dynamics.xml", # "newtrain-dynamic.xml", "preomr.xml"] files = ["preomr.xml", "preomr_edited.xml", "preomr_edited_cnn.xml"] import os.path # try to match with different trainingsets. for dynamic in ([d for d in files if os.path.isfile(d)]): ci.load_new_training_data(dynamic) print "%s - count_of_training=%d, k=%d" % (dynamic, len(c.stats), c.k) result = {} # Push into buckets based on the count of found glyphs. csv = {} sys.stdout.flush() # Try with different epsilon for false_positives: e_fp for e_fp in arange(0.01, 1.01, 0.01): c.e_fp = e_fp count = len(ci.classified_glyphs()) # Init bucket. if not result.has_key(count): result[count] = [] result[count].append((e_fp, c.d_t())) csv[e_fp] = count # Find the best match to the wanted result. k, res, diff = find_nearest(result, expected_count) confid = [(len(v), key, v[0][0], v[0][1]) for key, v in result.iteritems()] confid2 = [(key, len(v)) for key, v in result.iteritems()] confid.sort(reverse=True) confid2.sort() print "efp,count" for e_fp, c in sorted(csv.iteritems()): print "%s,%s" % (e_fp, c) print print "count,spansize" for count, spansize in confid2: print "%s,%s" % (count, spansize) return ret = [] for i in range(0, min(10, len(confid)) + 1): ret.append(confid[i]) if not result.has_key(expected_count): print "Never found the desired amount with %s" % dynamic print "Found in %d(%d): %s" % (k, diff, [r for r in res]) rgbimg = ci.image.to_rgb() cg = ci.classified_glyphs(res[0].d_t) [outline(rgbimg, g, 3.0, RGBPixel(255, 0, 0)) for g in cg] rgbimg.save_PNG("class_%s_%s.png" % (filename, dynamic)) print