// combining several vocab trees into one
int combineVocab(vector<string> &allTreesIn, char *tree_out)
{
int num_trees = (int)allTreesIn.size();
VocabTree tree;
tree.Read(allTreesIn[0].c_str());
//Start with the second tree, as we just read the first one
for (int i = 1; i < num_trees; i++)
{
printf("[VocabCombine] Adding tree %d [%s]...\n", i, allTreesIn[i].c_str());
fflush(stdout);
VocabTree tree_add;
tree_add.Read(allTreesIn[i].c_str());
tree.Combine(tree_add);
tree_add.Clear();
}
//Now do the reweighting:
int total_num_db_images = tree.GetMaxDatabaseImageIndex() + 1;
printf("Total num_db_images: %d\n", total_num_db_images);
tree.ComputeTFIDFWeights(total_num_db_images);
tree.NormalizeDatabase(0, total_num_db_images);
tree.Write(tree_out);
//Write vectors to a file
tree.WriteDatabaseVectors("vectors_all.txt", 0, total_num_db_images);
return 0;
}
int main(int argc, char **argv)
{
if (argc < 3) {
printf("Usage: %s <tree1.in> <tree2.in> ... <tree.out>\n",
argv[0]);
return 1;
}
int num_trees = argc - 2;
char *tree_out = argv[argc-1];
VocabTree tree;
tree.Read(argv[1]);
/* Start with the second tree, as we just read the first one */
for (int i = 1; i < num_trees; i++) {
printf("[VocabCombine] Adding tree %d [%s]...\n", i, argv[i+1]);
fflush(stdout);
VocabTree tree_add;
tree_add.Read(argv[i+1]);
tree.Combine(tree_add);
tree_add.Clear();
}
/* Now do the reweighting */
// if (use_tfidf)
int total_num_db_images = tree.GetMaxDatabaseImageIndex() + 1;
printf("Total num_db_images: %d\n", total_num_db_images);
tree.ComputeTFIDFWeights(total_num_db_images);
tree.NormalizeDatabase(0, total_num_db_images);
tree.Write(tree_out);
/* Write vectors to a file */
tree.WriteDatabaseVectors("vectors_all.txt", 0, total_num_db_images);
return 0;
}
int main(int argc, char **argv)
{
if (argc < 4 || argc > 8) {
printf("Usage: %s <list.in> <tree.in> <tree.out> [use_tfidf:1] "
"[normalize:1] [start_id:0] [distance_type:1]\n",
argv[0]);
return 1;
}
double min_feature_scale = 1.4;
bool use_tfidf = true;
bool normalize = true;
char *list_in = argv[1];
char *tree_in = argv[2];
char *tree_out = argv[3];
DistanceType distance_type = DistanceMin;
int start_id = 0;
if (argc >= 5)
use_tfidf = atoi(argv[4]);
if (argc >= 6)
normalize = atoi(argv[5]);
if (argc >= 7)
start_id = atoi(argv[6]);
if (argc >= 8)
distance_type = (DistanceType) atoi(argv[7]);
switch (distance_type) {
case DistanceDot:
printf("[VocabMatch] Using distance Dot\n");
break;
case DistanceMin:
printf("[VocabMatch] Using distance Min\n");
break;
default:
printf("[VocabMatch] Using no known distance!\n");
break;
}
FILE *f = fopen(list_in, "r");
if (f == NULL) {
printf("Error opening file %s for reading\n", list_in);
return 1;
}
std::vector<std::string> key_files;
char buf[256];
while (fgets(buf, 256, f)) {
/* Remove trailing newline */
if (buf[strlen(buf) - 1] == '\n')
buf[strlen(buf) - 1] = 0;
key_files.push_back(std::string(buf));
}
printf("[VocabBuildDB] Reading tree %s...\n", tree_in);
fflush(stdout);
VocabTree tree;
tree.Read(tree_in);
#if 1
tree.Flatten();
#endif
tree.m_distance_type = distance_type;
tree.SetInteriorNodeWeight(0.0);
/* Initialize leaf weights to 1.0 */
tree.SetConstantLeafWeights();
const int dim = 128;
int num_db_images = (int) key_files.size();
unsigned long count = 0;
tree.ClearDatabase();
for (int i = 0; i < num_db_images; i++) {
int num_keys = 0;
unsigned char *keys = ReadAndFilterKeys(key_files[i].c_str(),
dim, min_feature_scale,
0, num_keys);
printf("[VocabBuildDB] Adding vector %d (%d keys)\n",
start_id + i, num_keys);
tree.AddImageToDatabase(start_id + i, num_keys, keys);
if (num_keys > 0)
delete [] keys;
}
printf("[VocabBuildDB] Pushed %lu features\n", count);
fflush(stdout);
if (use_tfidf)
tree.ComputeTFIDFWeights(num_db_images);
if (normalize)
tree.NormalizeDatabase(start_id, num_db_images);
printf("[VocabBuildDB] Writing tree...\n");
tree.Write(tree_out);
// char filename[256];
// sprintf(filename, "vectors_%03d.txt", start_id);
// tree.WriteDatabaseVectors(filename, start_id, num_db_images);
return 0;
}
int main(int argc, char **argv)
{
if (argc != 6) {
printf("Usage: %s <list.in> <depth> <branching_factor> "
"<restarts> <tree.out>\n", argv[0]);
return 1;
}
const char *list_in = argv[1];
int depth = atoi(argv[2]);
int bf = atoi(argv[3]);
int restarts = atoi(argv[4]);
const char *tree_out = argv[5];
std::ofstream log("debug_VocabLearn.txt");
log << "Building tree with depth: " << depth << ", branching factor: " << bf << ", and restarts: " << restarts << std::endl;
printf("Building tree with depth: %d, branching factor: %d, "
"and restarts: %d\n", depth, bf, restarts);
FILE *f = fopen(list_in, "r");
std::vector<std::string> key_files;
char buf[256];
while (fgets(buf, 256, f)) {
/* Remove trailing newline */
if (buf[strlen(buf) - 1] == '\n')
buf[strlen(buf) - 1] = 0;
key_files.push_back(std::string(buf));
}
fclose(f);
int num_files = (int) key_files.size();
unsigned long total_keys = 0;
for (int i = 0; i < num_files; i++) {
int num_keys = GetNumberOfKeys(key_files[i].c_str());
total_keys += num_keys;
}
log << "Total number of keys: " << total_keys << std::endl;
printf("Total number of keys: %lu\n", total_keys);
fflush(stdout);
int dim = 128;
unsigned long long len = (unsigned long long) total_keys * dim;
int num_arrays =
len / MAX_ARRAY_SIZE + ((len % MAX_ARRAY_SIZE) == 0 ? 0 : 1);
unsigned char **vs = new unsigned char *[num_arrays];
log << "Allocating " << len << " bytes in total, in " << num_arrays << " arrays" << std::endl;
printf("Allocating %llu bytes in total, in %d arrays\n", len, num_arrays);
unsigned long long total = 0;
for (int i = 0; i < num_arrays; i++) {
unsigned long long remainder = len - total;
unsigned long len_curr = MIN(remainder, MAX_ARRAY_SIZE);
log << "Allocating array of size " << len_curr << std::endl;
printf("Allocating array of size %lu\n", len_curr);
fflush(stdout);
vs[i] = new unsigned char[len_curr];
remainder -= len_curr;
}
/* Create the array of pointers */
log << "Allocating pointer array of size " << sizeof(unsigned char) * total_keys << std::endl;
printf("Allocating pointer array of size %lu\n", 4 * total_keys);
fflush(stdout);
unsigned char **vp = new unsigned char *[total_keys];
unsigned long off = 0;
unsigned long curr_key = 0;
int curr_array = 0;
for (int i = 0; i < num_files; i++) {
log << "Reading key file " << key_files[i] << std::endl;
//printf(" Reading keyfile %s\n", key_files[i].c_str());
fflush(stdout);
unsigned char *keys;
int num_keys = 0;
keypt_t *info = NULL;
num_keys = ReadKeyFileCXX(key_files[i].c_str(), &keys);
log << "There are " << num_keys << " keys in file " << key_files[i] << std::endl;
// num_keys = ReadKeyFile(key_files[i].c_str(), &keys);
#ifdef DEBUG
std::ofstream debug_log("debug_log_2.txt");
debug_log << num_keys << " " << dim << std::endl;
for (int i=0; i<num_keys; i++) {
for (int k=0; k<dim; k++) {
debug_log << (int) keys[dim*i + k] << " ";
}
debug_log << std::endl;
}
debug_log.close();
#endif
if (num_keys > 0) {
for (int j = 0; j < num_keys; j++) {
#ifdef DEBUG
log << "Copying key " << j << std::endl;
#endif
for (int k = 0; k < dim; k++) {
#ifdef DEBUG
log << "\tReading ";
#endif
unsigned char val = keys[j * dim + k];
#ifdef DEBUG
log << (int) val << ". Writing ";
#endif
vs[curr_array][off + k] = val;
#ifdef DEBUG
log << (int) vs[curr_array][off + k] << std::endl;
#endif
}
#ifdef DEBUG
log << "vp[curr_key] = vs[curr_array] + off;" << std::endl;
#endif
vp[curr_key] = vs[curr_array] + off;
curr_key++;
off += dim;
if (off == MAX_ARRAY_SIZE) {
off = 0;
curr_array++;
}
}
delete [] keys;
if (info != NULL)
delete [] info;
}
}
log << "Building tree..." << std::endl;
VocabTree tree;
tree.Build(total_keys, dim, depth, bf, restarts, vp);
tree.Write(tree_out);
log.close();
return 0;
}
int main(int argc, char **argv)
{
if (argc != 6) {
printf("Usage: %s <list.in> <depth> <branching_factor> "
"<restarts> <tree.out>\n", argv[0]);
return 1;
}
const char *list_in = argv[1];
int depth = atoi(argv[2]);
int bf = atoi(argv[3]);
int restarts = atoi(argv[4]);
const char *tree_out = argv[5];
printf("Building tree with depth: %d, branching factor: %d, "
"and restarts: %d\n", depth, bf, restarts);
FILE *f = fopen(list_in, "r");
if (f == NULL) {
printf("Could not open file: %s\n", list_in);
return 1;
}
std::vector<std::string> key_files;
char buf[256];
while (fgets(buf, 256, f)) {
/* Remove trailing newline */
if (buf[strlen(buf) - 1] == '\n')
buf[strlen(buf) - 1] = 0;
key_files.push_back(std::string(buf));
}
fclose(f);
int num_files = (int) key_files.size();
unsigned long total_keys = 0;
for (int i = 0; i < num_files; i++) {
int num_keys = GetNumberOfKeys(key_files[i].c_str());
total_keys += num_keys;
}
printf("Total number of keys: %lu\n", total_keys);
// Reduce the branching factor if need be if there are not
// enough keys, to avoid problems later.
if (bf >= (int)total_keys){
bf = total_keys - 1;
printf("Reducing the branching factor to: %d\n", bf);
}
fflush(stdout);
int dim = 128;
unsigned long long len = (unsigned long long) total_keys * dim;
int num_arrays =
len / MAX_ARRAY_SIZE + ((len % MAX_ARRAY_SIZE) == 0 ? 0 : 1);
unsigned char **vs = new unsigned char *[num_arrays];
printf("Allocating %llu bytes in total, in %d arrays\n", len, num_arrays);
unsigned long long total = 0;
for (int i = 0; i < num_arrays; i++) {
unsigned long long remainder = len - total;
unsigned long len_curr = MIN(remainder, MAX_ARRAY_SIZE);
printf("Allocating array of size %lu\n", len_curr);
fflush(stdout);
vs[i] = new unsigned char[len_curr];
remainder -= len_curr;
}
/* Create the array of pointers */
printf("Allocating pointer array of size %lu\n", 4 * total_keys);
fflush(stdout);
unsigned char **vp = new unsigned char *[total_keys];
unsigned long off = 0;
unsigned long curr_key = 0;
int curr_array = 0;
for (int i = 0; i < num_files; i++) {
printf(" Reading keyfile %s\n", key_files[i].c_str());
fflush(stdout);
short int *keys;
int num_keys = 0;
keypt_t *info = NULL;
num_keys = ReadKeyFile(key_files[i].c_str(), &keys);
if (num_keys > 0) {
for (int j = 0; j < num_keys; j++) {
for (int k = 0; k < dim; k++) {
vs[curr_array][off + k] = keys[j * dim + k];
}
vp[curr_key] = vs[curr_array] + off;
curr_key++;
off += dim;
if (off == MAX_ARRAY_SIZE) {
off = 0;
curr_array++;
}
}
delete [] keys;
if (info != NULL)
delete [] info;
}
}
VocabTree tree;
tree.Build(total_keys, dim, depth, bf, restarts, vp);
tree.Write(tree_out);
return 0;
}
// Building a database with a vocabulary tree
int VocabDB(char *list_in, char *tree_in, char *db_out, int use_tfidf = 1, int normalize = 1, int start_id = 0, DistanceType distance_type = DistanceMin)
{
double min_feature_scale = 1.4;
switch (distance_type) {
case DistanceDot:
printf("[VocabMatch] Using distance Dot\n");
break;
case DistanceMin:
printf("[VocabMatch] Using distance Min\n");
break;
default:
printf("[VocabMatch] Using no known distance!\n");
break;
}
FILE *f = fopen(list_in, "r");
if (f == NULL) {
printf("Error opening file %s for reading\n", list_in);
return 1;
}
std::vector<std::string> key_files;
char buf[256];
while (fgets(buf, 256, f))
{
/* Remove trailing newline */
if (buf[strlen(buf) - 1] == '\n')
buf[strlen(buf) - 1] = 0;
key_files.push_back(std::string(buf));
}
printf("[VocabBuildDB] Reading tree %s...\n", tree_in);
fflush(stdout);
VocabTree tree;
tree.Read(tree_in);
tree.Flatten();
tree.m_distance_type = distance_type;
tree.SetInteriorNodeWeight(0.0);
//Initialize leaf weights to 1.0
tree.SetConstantLeafWeights();
const int dim = 128;
int num_db_images = (int)key_files.size();
unsigned long count = 0;
tree.ClearDatabase();
for (int i = 0; i < num_db_images; i++)
{
int num_keys = 0;
unsigned char *keys = ReadAndFilterKeys(key_files[i].c_str(), dim, min_feature_scale, 0, num_keys);
printf("[VocabBuildDB] Adding vector %d (%d keys)\n", start_id + i, num_keys);
tree.AddImageToDatabase(start_id + i, num_keys, keys);
if (num_keys > 0)
delete[] keys;
}
printf("[VocabBuildDB] Pushed %lu features\n", count);
fflush(stdout);
if (use_tfidf == 1)
tree.ComputeTFIDFWeights(num_db_images);
if (normalize == 1)
tree.NormalizeDatabase(start_id, num_db_images);
printf("[VocabBuildDB] Writing database ...\n");
tree.Write(db_out);
// char filename[256];
// sprintf(filename, "vectors_%03d.txt", start_id);
// tree.WriteDatabaseVectors(filename, start_id, num_db_images);
return 0;
}