Added init_cpu() function to init CPU just to speedup the 1st detection (if GPU=0)

build/darknet/x64/darknet.py

@@ -144,6 +144,8 @@ if hasGPU:
     set_gpu = lib.cuda_set_device
     set_gpu.argtypes = [c_int]
 
+init_cpu = lib.init_cpu
+
 make_image = lib.make_image
 make_image.argtypes = [c_int, c_int, c_int]
 make_image.restype = IMAGE

darknet.py

@@ -144,6 +144,8 @@ if hasGPU:
     set_gpu = lib.cuda_set_device
     set_gpu.argtypes = [c_int]
 
+init_cpu = lib.init_cpu
+
 make_image = lib.make_image
 make_image.argtypes = [c_int, c_int, c_int]
 make_image.restype = IMAGE

include/darknet.h

@@ -921,6 +921,9 @@ void stop_timer_and_show();
 void stop_timer_and_show_name(char *name);
 void show_total_time();
 
+// gemm.h
+LIB_API void init_cpu();
+
 #ifdef __cplusplus
 }
 #endif  // __cplusplus

src/gemm.c

@@ -2855,3 +2855,10 @@ int test_gpu_blas()
     return 0;
 }
 #endif
+
+
+
+void init_cpu() {
+    is_avx();
+    is_fma_avx2();
+}

src/parser.c

@@ -379,7 +379,7 @@ layer parse_yolo(list *options, size_params params)
         iou_loss, l.iou_loss, l.iou_normalizer, l.cls_normalizer, l.scale_x_y);
 
     l.beta_nms = option_find_float_quiet(options, "beta_nms", 0.6);
-    char *nms_kind = option_find_str(options, "nms_kind", "default");
+    char *nms_kind = option_find_str_quiet(options, "nms_kind", "default");
     if (strcmp(nms_kind, "default") == 0) l.nms_kind = DEFAULT_NMS;
     else {
         if (strcmp(nms_kind, "greedynms") == 0) l.nms_kind = GREEDY_NMS;
@@ -469,7 +469,7 @@ layer parse_gaussian_yolo(list *options, size_params params) // Gaussian_YOLOv3
     else l.iou_loss = IOU;
 
     l.beta_nms = option_find_float_quiet(options, "beta_nms", 0.6);
-    char *nms_kind = option_find_str(options, "nms_kind", "default");
+    char *nms_kind = option_find_str_quiet(options, "nms_kind", "default");
     if (strcmp(nms_kind, "default") == 0) l.nms_kind = DEFAULT_NMS;
     else {
         if (strcmp(nms_kind, "greedynms") == 0) l.nms_kind = GREEDY_NMS;