From 028696bf15efeca3acb3db8c42a96f7b9e0f55ff Mon Sep 17 00:00:00 2001
From: iovodov <b@ovdv.ru>
Date: Thu, 03 May 2018 13:33:46 +0000
Subject: [PATCH] Output improvements for detector results: When printing detector results, output was done in random order, obfuscating results for interpreting. Now: 1. Text output includes coordinates of rects in (left,right,top,bottom in pixels) along with label and score 2. Text output is sorted by rect lefts to simplify finding appropriate rects on image 3. If several class probs are > thresh for some detection, the most probable is written first and coordinates for others are not repeated 4. Rects are imprinted in image in order by their best class prob, so most probable rects are always on top and not overlayed by less probable ones 5. Most probable label for rect is always written first Also: 6. Message about low GPU memory include required amount
---
src/crop_layer_kernels.cu | 25 +++++++++++++++++--------
1 files changed, 17 insertions(+), 8 deletions(-)
diff --git a/src/crop_layer_kernels.cu b/src/crop_layer_kernels.cu
index fc7fcbd..8a08630 100644
--- a/src/crop_layer_kernels.cu
+++ b/src/crop_layer_kernels.cu
@@ -1,3 +1,7 @@
+#include "cuda_runtime.h"
+#include "curand.h"
+#include "cublas_v2.h"
+
extern "C" {
#include "crop_layer.h"
#include "utils.h"
@@ -91,7 +95,7 @@
return val;
}
-__global__ void levels_image_kernel(float *image, float *rand, int batch, int w, int h, int train, float saturation, float exposure, float translate, float scale)
+__global__ void levels_image_kernel(float *image, float *rand, int batch, int w, int h, int train, float saturation, float exposure, float translate, float scale, float shift)
{
int size = batch * w * h;
int id = (blockIdx.x + blockIdx.y*gridDim.x) * blockDim.x + threadIdx.x;
@@ -100,6 +104,9 @@
id /= w;
int y = id % h;
id /= h;
+ float rshift = rand[0];
+ float gshift = rand[1];
+ float bshift = rand[2];
float r0 = rand[8*id + 0];
float r1 = rand[8*id + 1];
float r2 = rand[8*id + 2];
@@ -121,10 +128,12 @@
hsv.y *= saturation;
hsv.z *= exposure;
rgb = hsv_to_rgb_kernel(hsv);
+ } else {
+ shift = 0;
}
- image[x + w*(y + h*0)] = rgb.x*scale + translate;
- image[x + w*(y + h*1)] = rgb.y*scale + translate;
- image[x + w*(y + h*2)] = rgb.z*scale + translate;
+ image[x + w*(y + h*0)] = rgb.x*scale + translate + (rshift - .5)*shift;
+ image[x + w*(y + h*1)] = rgb.y*scale + translate + (gshift - .5)*shift;
+ image[x + w*(y + h*2)] = rgb.z*scale + translate + (bshift - .5)*shift;
}
__global__ void forward_crop_layer_kernel(float *input, float *rand, int size, int c, int h, int w, int crop_height, int crop_width, int train, int flip, float angle, float *output)
@@ -175,7 +184,7 @@
{
cuda_random(layer.rand_gpu, layer.batch*8);
- float radians = layer.angle*3.14159/180.;
+ float radians = layer.angle*3.14159265/180.;
float scale = 2;
float translate = -1;
@@ -186,12 +195,12 @@
int size = layer.batch * layer.w * layer.h;
- levels_image_kernel<<<cuda_gridsize(size), BLOCK>>>(state.input, layer.rand_gpu, layer.batch, layer.w, layer.h, state.train, layer.saturation, layer.exposure, translate, scale);
+ levels_image_kernel<<<cuda_gridsize(size), BLOCK>>>(state.input, layer.rand_gpu, layer.batch, layer.w, layer.h, state.train, layer.saturation, layer.exposure, translate, scale, layer.shift);
check_error(cudaPeekAtLastError());
- size = layer.batch*layer.c*layer.crop_width*layer.crop_height;
+ size = layer.batch*layer.c*layer.out_w*layer.out_h;
- forward_crop_layer_kernel<<<cuda_gridsize(size), BLOCK>>>(state.input, layer.rand_gpu, size, layer.c, layer.h, layer.w, layer.crop_height, layer.crop_width, state.train, layer.flip, radians, layer.output_gpu);
+ forward_crop_layer_kernel<<<cuda_gridsize(size), BLOCK>>>(state.input, layer.rand_gpu, size, layer.c, layer.h, layer.w, layer.out_h, layer.out_w, state.train, layer.flip, radians, layer.output_gpu);
check_error(cudaPeekAtLastError());
/*
--
Gitblit v1.10.0