Experimental reinforcement learning.

AlexeyAB

2018-05-22 032acd408acd119a5bd5c132765fcdd6caedea6b

 src/detector.c

@@ -542,6 +542,7 @@
   char *mapf = option_find_str(options, "map", 0);
   int *map = 0;
   if (mapf) map = read_map(mapf);
   FILE* reinforcement_fd = NULL;

   network net = parse_network_cfg_custom(cfgfile, 1);   // set batch=1
   if (weightfile) {
@@ -723,6 +724,15 @@
         
         unique_truth_count += num_labels;

         //static int previous_errors = 0;
         //int total_errors = fp_for_thresh + (unique_truth_count - tp_for_thresh);
         //int errors_in_this_image = total_errors - previous_errors;
         //previous_errors = total_errors;
         //if(reinforcement_fd == NULL) reinforcement_fd = fopen("reinforcement.txt", "wb");
         //char buff[1000];
         //sprintf(buff, "%s\n", path);
         //if(errors_in_this_image > 0) fwrite(buff, sizeof(char), strlen(buff), reinforcement_fd);

         free_detections(dets, nboxes);
         free(id);
         free_image(val[t]);
@@ -843,6 +853,7 @@
   free(truth_classes_count);

   fprintf(stderr, "Total Detection Time: %f Seconds\n", (double)(time(0) - start));
   if (reinforcement_fd != NULL) fclose(reinforcement_fd);
}

#ifdef OPENCV