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/layer.h | 170 +++++++++++++++++++++++++++++++++++++++++++++++++++-----
1 files changed, 155 insertions(+), 15 deletions(-)
diff --git a/src/layer.h b/src/layer.h
index 1b12009..81e27ad 100644
--- a/src/layer.h
+++ b/src/layer.h
@@ -2,6 +2,10 @@
#define BASE_LAYER_H
#include "activations.h"
+#include "stddef.h"
+#include "tree.h"
+
+struct network_state;
struct layer;
typedef struct layer layer;
@@ -20,18 +24,50 @@
NORMALIZATION,
AVGPOOL,
LOCAL,
- SHORTCUT
+ SHORTCUT,
+ ACTIVE,
+ RNN,
+ GRU,
+ CRNN,
+ BATCHNORM,
+ NETWORK,
+ XNOR,
+ REGION,
+ YOLO,
+ REORG,
+ UPSAMPLE,
+ REORG_OLD,
+ BLANK
} LAYER_TYPE;
typedef enum{
- SSE, MASKED
+ SSE, MASKED, SMOOTH
} COST_TYPE;
+typedef struct {
+ int batch;
+ float learning_rate;
+ float momentum;
+ float decay;
+ int adam;
+ float B1;
+ float B2;
+ float eps;
+ int t;
+} update_args;
+
struct layer{
LAYER_TYPE type;
ACTIVATION activation;
COST_TYPE cost_type;
+ void (*forward) (struct layer, struct network_state);
+ void (*backward) (struct layer, struct network_state);
+ void (*update) (struct layer, int, float, float, float);
+ void (*forward_gpu) (struct layer, struct network_state);
+ void (*backward_gpu) (struct layer, struct network_state);
+ void (*update_gpu) (struct layer, int, float, float, float);
int batch_normalize;
+ int shortcut;
int batch;
int forced;
int flipped;
@@ -41,21 +77,28 @@
int h,w,c;
int out_h, out_w, out_c;
int n;
+ int max_boxes;
int groups;
int size;
int side;
int stride;
+ int reverse;
int pad;
- int crop_width;
- int crop_height;
int sqrt;
int flip;
int index;
+ int binary;
+ int xnor;
+ int steps;
+ int hidden;
+ float dot;
float angle;
float jitter;
float saturation;
float exposure;
float shift;
+ float ratio;
+ int focal_loss;
int softmax;
int classes;
int coords;
@@ -65,6 +108,24 @@
int does_cost;
int joint;
int noadjust;
+ int reorg;
+ int log;
+ int tanh;
+ int *mask;
+ int total;
+
+ int adam;
+ float B1;
+ float B2;
+ float eps;
+ float *m_gpu;
+ float *v_gpu;
+ int t;
+ float *m;
+ float *v;
+
+ tree *softmax_tree;
+ int *map;
float alpha;
float beta;
@@ -73,19 +134,40 @@
float coord_scale;
float object_scale;
float noobject_scale;
+ float mask_scale;
float class_scale;
+ int bias_match;
+ int random;
+ float ignore_thresh;
+ float truth_thresh;
+ float thresh;
+ float focus;
+ int classfix;
+ int absolute;
+ int onlyforward;
+ int stopbackward;
int dontload;
int dontloadscales;
+ float temperature;
float probability;
float scale;
int *indexes;
float *rand;
float *cost;
- float *filters;
- float *filter_updates;
+ char *cweights;
+ float *state;
+ float *prev_state;
+ float *forgot_state;
+ float *forgot_delta;
+ float *state_delta;
+
+ float *concat;
+ float *concat_delta;
+
+ float *binary_weights;
float *biases;
float *bias_updates;
@@ -108,16 +190,63 @@
float * mean;
float * variance;
+ float * mean_delta;
+ float * variance_delta;
+
float * rolling_mean;
float * rolling_variance;
- #ifdef GPU
- int *indexes_gpu;
- float * filters_gpu;
- float * filter_updates_gpu;
+ float * x;
+ float * x_norm;
- float * spatial_mean_gpu;
- float * spatial_variance_gpu;
+ struct layer *input_layer;
+ struct layer *self_layer;
+ struct layer *output_layer;
+
+ struct layer *input_gate_layer;
+ struct layer *state_gate_layer;
+ struct layer *input_save_layer;
+ struct layer *state_save_layer;
+ struct layer *input_state_layer;
+ struct layer *state_state_layer;
+
+ struct layer *input_z_layer;
+ struct layer *state_z_layer;
+
+ struct layer *input_r_layer;
+ struct layer *state_r_layer;
+
+ struct layer *input_h_layer;
+ struct layer *state_h_layer;
+
+ float *z_cpu;
+ float *r_cpu;
+ float *h_cpu;
+
+ float *binary_input;
+
+ size_t workspace_size;
+
+ #ifdef GPU
+ float *z_gpu;
+ float *r_gpu;
+ float *h_gpu;
+
+ int *indexes_gpu;
+ float * prev_state_gpu;
+ float * forgot_state_gpu;
+ float * forgot_delta_gpu;
+ float * state_gpu;
+ float * state_delta_gpu;
+ float * gate_gpu;
+ float * gate_delta_gpu;
+ float * save_gpu;
+ float * save_delta_gpu;
+ float * concat_gpu;
+ float * concat_delta_gpu;
+
+ float *binary_input_gpu;
+ float *binary_weights_gpu;
float * mean_gpu;
float * variance_gpu;
@@ -125,9 +254,6 @@
float * rolling_mean_gpu;
float * rolling_variance_gpu;
- float * spatial_mean_delta_gpu;
- float * spatial_variance_delta_gpu;
-
float * variance_delta_gpu;
float * mean_delta_gpu;
@@ -138,6 +264,9 @@
float * weights_gpu;
float * weight_updates_gpu;
+ float * weights_gpu16;
+ float * weight_updates_gpu16;
+
float * biases_gpu;
float * bias_updates_gpu;
@@ -149,6 +278,17 @@
float * rand_gpu;
float * squared_gpu;
float * norms_gpu;
+ #ifdef CUDNN
+ cudnnTensorDescriptor_t srcTensorDesc, dstTensorDesc;
+ cudnnTensorDescriptor_t dsrcTensorDesc, ddstTensorDesc;
+ cudnnTensorDescriptor_t normTensorDesc, normDstTensorDesc, normDstTensorDescF16;
+ cudnnFilterDescriptor_t weightDesc;
+ cudnnFilterDescriptor_t dweightDesc;
+ cudnnConvolutionDescriptor_t convDesc;
+ cudnnConvolutionFwdAlgo_t fw_algo;
+ cudnnConvolutionBwdDataAlgo_t bd_algo;
+ cudnnConvolutionBwdFilterAlgo_t bf_algo;
+ #endif
#endif
};
--
Gitblit v1.10.0