avfilter/dnn/dnn_backend_tf: simplify the code with ff_hex_to_data

please use tools/python/tf_sess_config.py to get the sess_config after that. note the byte order of session config is in normal order. bump the MICRO version for the config change.

Limin Wang committed 5 years ago

Commit f183d651 parent 4e8d2243 files · +54 −35Browse files

Unified · Split

3 files

libavfilter/dnn/dnn_backend_tf.c +8 −34
libavfilter/version.h +1 −1
tools/python/tf_sess_config.py +45 −0

libavfilter/dnn/dnn_backend_tf.c+8−34

@@ -28,6 +28,7 @@ · 27 unmodified lines
28	#include "dnn_backend_native_layer_conv2d.h"	28	#include "dnn_backend_native_layer_conv2d.h"
29	#include "dnn_backend_native_layer_depth2space.h"	29	#include "dnn_backend_native_layer_depth2space.h"
30	#include "libavformat/avio.h"	30	#include "libavformat/avio.h"
		31	#include "libavformat/internal.h"
31	#include "libavutil/avassert.h"	32	#include "libavutil/avassert.h"
32	#include "../internal.h"	33	#include "../internal.h"
33	#include "dnn_backend_native_layer_pad.h"	34	#include "dnn_backend_native_layer_pad.h"
@@ -206,53 +207,26 @@ static DNNReturnType load_tf_model(TFModel tf_model, const char model_filename · 172 unmodified lines
206		207
207	// prepare the sess config data	208	// prepare the sess config data
208	if (tf_model->ctx.options.sess_config != NULL) {	209	if (tf_model->ctx.options.sess_config != NULL) {
		210	const char *config;
209	/*	211	/*
210	tf_model->ctx.options.sess_config is hex to present the serialized proto	212	tf_model->ctx.options.sess_config is hex to present the serialized proto
211	required by TF_SetConfig below, so we need to first generate the serialized	213	required by TF_SetConfig below, so we need to first generate the serialized
212	proto in a python script, the following is a script example to generate	214	proto in a python script, tools/python/tf_sess_config.py is a script example
213	serialized proto which specifies one GPU, we can change the script to add	215	to generate the configs of sess_config.
214	more options.
215
216	import tensorflow as tf
217	gpu_options = tf.GPUOptions(visible_device_list='0')
218	config = tf.ConfigProto(gpu_options=gpu_options)
219	s = config.SerializeToString()
220	b = ''.join("%02x" % int(ord(b)) for b in s[::-1])
221	print('0x%s' % b)
222
223	the script output looks like: 0xab...cd, and then pass 0xab...cd to sess_config.
224	*/	216	*/
225	char tmp[3];
226	tmp[2] = '\0';
227
228	if (strncmp(tf_model->ctx.options.sess_config, "0x", 2) != 0) {	217	if (strncmp(tf_model->ctx.options.sess_config, "0x", 2) != 0) {
229	av_log(ctx, AV_LOG_ERROR, "sess_config should start with '0x'\n");	218	av_log(ctx, AV_LOG_ERROR, "sess_config should start with '0x'\n");
230	return DNN_ERROR;	219	return DNN_ERROR;
231	}	220	}
		221	config = tf_model->ctx.options.sess_config + 2;
		222	sess_config_length = ff_hex_to_data(NULL, config);
232		223
233	sess_config_length = strlen(tf_model->ctx.options.sess_config);	224	sess_config = av_mallocz(sess_config_length + AV_INPUT_BUFFER_PADDING_SIZE);
234	if (sess_config_length % 2 != 0) {
235	av_log(ctx, AV_LOG_ERROR, "the length of sess_config is not even (%s), "
236	"please re-generate the config.\n",
237	tf_model->ctx.options.sess_config);
238	return DNN_ERROR;
239	}
240
241	sess_config_length -= 2; //ignore the first '0x'
242	sess_config_length /= 2; //get the data length in byte
243
244	sess_config = av_malloc(sess_config_length);
245	if (!sess_config) {	225	if (!sess_config) {
246	av_log(ctx, AV_LOG_ERROR, "failed to allocate memory\n");	226	av_log(ctx, AV_LOG_ERROR, "failed to allocate memory\n");
247	return DNN_ERROR;	227	return DNN_ERROR;
248	}	228	}
249		229	ff_hex_to_data(sess_config, config);
250	for (int i = 0; i < sess_config_length; i++) {
251	int index = 2 + (sess_config_length - 1 - i) * 2;
252	tmp[0] = tf_model->ctx.options.sess_config[index];
253	tmp[1] = tf_model->ctx.options.sess_config[index + 1];
254	sess_config[i] = strtol(tmp, NULL, 16);
255	}
256	}	230	}
257		231
258	graph_def = read_graph(model_filename);	232	graph_def = read_graph(model_filename);

libavfilter/version.h+1−1

@@ -31,7 +31,7 @@ · 30 unmodified lines
31		31
32	#define LIBAVFILTER_VERSION_MAJOR 8	32	#define LIBAVFILTER_VERSION_MAJOR 8
33	#define LIBAVFILTER_VERSION_MINOR 0	33	#define LIBAVFILTER_VERSION_MINOR 0
34	#define LIBAVFILTER_VERSION_MICRO 100	34	#define LIBAVFILTER_VERSION_MICRO 101
35		35
36		36
37	#define LIBAVFILTER_VERSION_INT AV_VERSION_INT(LIBAVFILTER_VERSION_MAJOR, \	37	#define LIBAVFILTER_VERSION_INT AV_VERSION_INT(LIBAVFILTER_VERSION_MAJOR, \

tools/python/tf_sess_config.py+45−0

@@ -0,0 +1,45 @@
	1	# Copyright (c) 2021
	2	#
	3	# This file is part of FFmpeg.
	4	#
	5	# FFmpeg is free software; you can redistribute it and/or
	6	# modify it under the terms of the GNU Lesser General Public
	7	# License as published by the Free Software Foundation; either
	8	# version 2.1 of the License, or (at your option) any later version.
	9	#
	10	# FFmpeg is distributed in the hope that it will be useful,
	11	# but WITHOUT ANY WARRANTY; without even the implied warranty of
	12	# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
	13	# Lesser General Public License for more details.
	14	#
	15	# You should have received a copy of the GNU Lesser General Public
	16	# License along with FFmpeg; if not, write to the Free Software
	17	# Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
	18	# ==============================================================================
	19
	20	# verified with Python 3.6.8 on CentOS 7.2
	21	import tensorflow as tf
	22
	23	visible_device_list = '0' # use , separator for more GPUs like '0, 1'
	24	per_process_gpu_memory_fraction = 0.9 # avoid out of memory
	25	intra_op_parallelism_threads = 2 # default in tensorflow
	26	inter_op_parallelism_threads = 5 # default in tensorflow
	27
	28	gpu_options = tf.compat.v1.GPUOptions(
	29	per_process_gpu_memory_fraction = per_process_gpu_memory_fraction,
	30	visible_device_list = visible_device_list,
	31	allow_growth = True)
	32
	33	config = tf.compat.v1.ConfigProto(
	34	allow_soft_placement = True,
	35	log_device_placement = False,
	36	intra_op_parallelism_threads = intra_op_parallelism_threads,
	37	inter_op_parallelism_threads = inter_op_parallelism_threads,
	38	gpu_options = gpu_options)
	39
	40	s = config.SerializeToString()
	41	# print(list(map(hex, s))) # print by json if need
	42
	43	print('a serialized protobuf string for TF_SetConfig, note the byte order is in normal order.')
	44	b = ''.join(format(b,'02x') for b in s)
	45	print('0x%s' % b) # print by hex format

avfilter/dnn/dnn_backend_tf: simplify the code with ff_hex_to_data

Keyboard shortcuts

General

Issues and pull requests

Issues, pull requests and Inbox

Repository code