open_toontown_panda3d/panda/src/grutil/ffmpegTexture.cxx

776 lines
25 KiB
C++

// Filename: ffmpegTexture.cxx
// Created by: zacpavlov (05May06)
//
////////////////////////////////////////////////////////////////////
//
// PANDA 3D SOFTWARE
// Copyright (c) Carnegie Mellon University. All rights reserved.
//
// All use of this software is subject to the terms of the revised BSD
// license. You should have received a copy of this license along
// with this source code in a file named "LICENSE."
//
////////////////////////////////////////////////////////////////////
#include "pandabase.h"
#ifdef HAVE_FFMPEG
#include "ffmpegTexture.h"
#include "clockObject.h"
#include "config_gobj.h"
#include "config_grutil.h"
#include "bamCacheRecord.h"
#include "bamReader.h"
TypeHandle FFMpegTexture::_type_handle;
////////////////////////////////////////////////////////////////////
// Function: FFMpegTexture::Constructor
// Access: Published
// Description: Sets up the texture to read frames from a camera
////////////////////////////////////////////////////////////////////
FFMpegTexture::
FFMpegTexture(const string &name) :
VideoTexture(name)
{
}
////////////////////////////////////////////////////////////////////
// Function: FFMpegTexture::Copy Constructor
// Access: Protected
// Description: Use FFmpegTexture::make_copy() to make a duplicate copy of
// an existing FFMpegTexture.
////////////////////////////////////////////////////////////////////
FFMpegTexture::
FFMpegTexture(const FFMpegTexture &copy) :
VideoTexture(copy),
_pages(copy._pages)
{
nassertv(false);
}
////////////////////////////////////////////////////////////////////
// Function: FFMpegTexture::Destructor
// Access: Published, Virtual
// Description: I'm betting that texture takes care of the, so we'll just do a clear.
////////////////////////////////////////////////////////////////////
FFMpegTexture::
~FFMpegTexture() {
clear();
}
////////////////////////////////////////////////////////////////////
// Function: FFMpegTexture::do_make_copy
// Access: Protected, Virtual
// Description: Returns a new copy of the same Texture. This copy,
// if applied to geometry, will be copied into texture
// as a separate texture from the original, so it will
// be duplicated in texture memory (and may be
// independently modified if desired).
//
// If the Texture is an FFMpegTexture, the resulting
// duplicate may be animated independently of the
// original.
////////////////////////////////////////////////////////////////////
PT(Texture) FFMpegTexture::
do_make_copy() {
PT(FFMpegTexture) tex = new FFMpegTexture(get_name());
tex->do_assign(*this);
return tex.p();
}
////////////////////////////////////////////////////////////////////
// Function: FFMpegTexture::do_assign
// Access: Protected
// Description: Implements make_copy().
////////////////////////////////////////////////////////////////////
void FFMpegTexture::
do_assign(const FFMpegTexture &copy) {
VideoTexture::do_assign(copy);
_pages = copy._pages;
}
////////////////////////////////////////////////////////////////////
// Function: FFMPegTexture::modify_page
// Access: Private
// Description: Returns a reference to the zth VideoPage (level) of
// the texture. In the case of a 2-d texture, there is
// only one page, level 0; but cube maps and 3-d
// textures have more.
////////////////////////////////////////////////////////////////////
FFMpegTexture::VideoPage &FFMpegTexture::
modify_page(int z) {
nassertr(z < _z_size, _pages[0]);
while (z >= (int)_pages.size()) {
_pages.push_back(VideoPage());
}
return _pages[z];
}
////////////////////////////////////////////////////////////////////
// Function: FFmpegTexture::do_reconsider_video_properties
// Access: Private
// Description: Resets the internal Texture properties when a new
// video file is loaded. Returns true if the new image
// is valid, false otherwise.
////////////////////////////////////////////////////////////////////
bool FFMpegTexture::
do_reconsider_video_properties(const FFMpegTexture::VideoStream &stream,
int num_components, int z,
const LoaderOptions &options) {
double frame_rate = 0.0f;
int num_frames = 0;
if (!stream._codec_context) {
// printf("not valid yet\n");
return true;
}
AVStream *vstream = stream._format_context->streams[stream._stream_number];
if (stream.is_from_file()) {
// frame rate comes from ffmpeg as an avRational.
frame_rate = vstream->r_frame_rate.num/(float)vstream->r_frame_rate.den;
// Number of frames is a little questionable if we've got variable
// frame rate. Duration comes in as a generic timestamp,
// and is therefore multiplied by AV_TIME_BASE.
num_frames = (int)((stream._format_context->duration*frame_rate)/AV_TIME_BASE);
if (grutil_cat.is_debug()) {
grutil_cat.debug()
<< "Loaded " << stream._filename << ", " << num_frames << " frames at "
<< frame_rate << " fps\n";
}
}
int width = stream._codec_context->width;
int height = stream._codec_context->height;
int x_size = width;
int y_size = height;
if (Texture::get_textures_power_2() != ATS_none) {
x_size = up_to_power_2(width);
y_size = up_to_power_2(height);
}
if (grutil_cat.is_debug()) {
grutil_cat.debug()
<< "Video stream is " << width << " by " << height
<< " pixels; fitting in texture " << x_size << " by "
<< y_size << " texels.\n";
}
if (!do_reconsider_image_properties(x_size, y_size, num_components,
T_unsigned_byte, z, options)) {
return false;
}
if (_loaded_from_image &&
(get_video_width() != width || get_video_height() != height ||
get_num_frames() != num_frames || get_frame_rate() != frame_rate)) {
grutil_cat.error()
<< "Video properties have changed for texture " << get_name()
<< " level " << z << ".\n";
return false;
}
set_frame_rate(frame_rate);
set_num_frames(num_frames);
set_video_size(width, height);
// By default, the newly-loaded video stream will immediately start
// looping.
loop(true);
return true;
}
////////////////////////////////////////////////////////////////////
// Function: FFMpegTexture::make_texture
// Access: Public, Static
// Description: A factory function to make a new FFMpegTexture, used
// to pass to the TexturePool.
////////////////////////////////////////////////////////////////////
PT(Texture) FFMpegTexture::
make_texture() {
return new FFMpegTexture;
}
////////////////////////////////////////////////////////////////////
// Function: FFMPegTexture::update_frame
// Access: Protected, Virtual
// Description: Called once per frame, as needed, to load the new
// image contents.
////////////////////////////////////////////////////////////////////
void FFMpegTexture::
update_frame(int frame) {
int max_z = max(_z_size, (int)_pages.size());
for (int z = 0; z < max_z; ++z) {
VideoPage &page = _pages[z];
if (page._color.is_valid() || page._alpha.is_valid()) {
modify_ram_image();
}
if (page._color.is_valid()) {
nassertv(get_num_components() >= 3 && get_component_width() == 1);
// A little different from the opencv implementation
// The frame is kept on the stream itself. This is partially
// because there is a conversion step that must be done for
// every video (I've gotten very odd results with any video
// that I don't convert, even if the IO formats are the same!)
if (page._color.get_frame_data(frame)) {
nassertv(get_video_width() <= _x_size && get_video_height() <= _y_size);
unsigned char *dest = _ram_images[0]._image.p() + get_expected_ram_page_size() * z;
int dest_row_width = (_x_size * _num_components * _component_width);
// Simplest case, where we deal with an rgb texture
if (get_num_components() == 3) {
int source_row_width=3*page._color._codec_context->width;
unsigned char * source=(unsigned char *)page._color._frame_out->data[0]
+source_row_width*(get_video_height()-1);
// row by row copy.
for (int y = 0; y < get_video_height(); ++y) {
memcpy(dest, source, source_row_width);
dest += dest_row_width;
source -= source_row_width;
}
// Next best option, we're a 4 component alpha video on one stream
} else if (page._color._codec_context->pix_fmt==PIX_FMT_RGBA32) {
int source_row_width= page._color._codec_context->width * 4;
unsigned char * source=(unsigned char *)page._color._frame_out->data[0]
+source_row_width*(get_video_height()-1);
// row by row copy.
for (int y = 0; y < get_video_height(); ++y) {
memcpy(dest,source,source_row_width);
dest += dest_row_width;
source -= source_row_width;
}
// Otherwise, we've got to be tricky
} else {
int source_row_width= page._color._codec_context->width * 3;
unsigned char * source=(unsigned char *)page._color._frame_out->data[0]
+source_row_width*(get_video_height()-1);
// The harder case--interleave the color in with the alpha,
// pixel by pixel.
nassertv(get_num_components() == 4);
for (int y = 0; y < get_video_height(); ++y) {
int dx = 0;
int sx = 0;
for (int x = 0; x < get_video_width(); ++x) {
dest[dx] = source[sx];
dest[dx + 1] = source[sx + 1];
dest[dx + 2] = source[sx + 2];
dx += 4;
sx += 3;
}
dest += dest_row_width;
source -= source_row_width;
}
}
}
}
if (page._alpha.is_valid()) {
nassertv(get_num_components() == 4 && get_component_width() == 1);
if (page._alpha.get_frame_data(frame)) {
nassertv(get_video_width() <= _x_size && get_video_height() <= _y_size);
// Currently, we assume the alpha has been converted to an rgb format
// There is no reason it can't be a 256 color grayscale though.
unsigned char *dest = _ram_images[0]._image.p() + get_expected_ram_page_size() * z;
int dest_row_width = (_x_size * _num_components * _component_width);
int source_row_width= page._alpha._codec_context->width * 3;
unsigned char * source=(unsigned char *)page._alpha._frame_out->data[0]
+source_row_width*(get_video_height()-1);
for (int y = 0; y < get_video_height(); ++y) {
int dx = 3;
int sx = 0;
for (int x = 0; x < get_video_width(); ++x) {
dest[dx] = source[sx];
dx += 4;
sx += 3;
}
dest += dest_row_width;
source -= source_row_width;
}
}
}
}
}
////////////////////////////////////////////////////////////////////
// Function: FFMpegTexture::do_read_one
// Access: Protected, Virtual
// Description: Combines a color and alpha video image from the two
// indicated filenames. Both must be the same kind of
// video with similar properties.
////////////////////////////////////////////////////////////////////
bool FFMpegTexture::
do_read_one(const Filename &fullpath, const Filename &alpha_fullpath,
int z, int n, int primary_file_num_channels, int alpha_file_channel,
const LoaderOptions &options,
bool header_only, BamCacheRecord *record) {
if (record != (BamCacheRecord *)NULL) {
record->add_dependent_file(fullpath);
}
nassertr(n == 0, false);
nassertr(z >= 0 && z < get_z_size(), false);
VideoPage &page = modify_page(z);
if (!page._color.read(fullpath)) {
grutil_cat.error()
<< "FFMpeg couldn't read " << fullpath << " as video.\n";
return false;
}
if (!alpha_fullpath.empty()) {
if (!page._alpha.read(alpha_fullpath)) {
grutil_cat.error()
<< "FFMPEG couldn't read " << alpha_fullpath << " as video.\n";
page._color.clear();
return false;
}
}
if (z == 0) {
if (!has_name()) {
set_name(fullpath.get_basename_wo_extension());
}
if (!has_filename()) {
_filename = fullpath;
_alpha_filename = alpha_fullpath;
}
_fullpath = fullpath;
_alpha_fullpath = alpha_fullpath;
}
if (page._color._codec_context->pix_fmt==PIX_FMT_RGBA32) {
// There had better not be an alpha interleave here.
nassertr(alpha_fullpath.empty(), false);
_primary_file_num_channels = 4;
_alpha_file_channel = 0;
if (!do_reconsider_video_properties(page._color, 4, z, options)) {
page._color.clear();
return false;
}
} else {
_primary_file_num_channels = 3;
_alpha_file_channel = alpha_file_channel;
if (page._alpha.is_valid()) {
if (!do_reconsider_video_properties(page._color, 4, z, options)) {
page._color.clear();
page._alpha.clear();
return false;
}
if (!do_reconsider_video_properties(page._alpha, 4, z, options)) {
page._color.clear();
page._alpha.clear();
return false;
}
} else {
if (!do_reconsider_video_properties(page._color, 3, z, options)) {
page._color.clear();
page._alpha.clear();
return false;
}
}
}
set_loaded_from_image();
clear_current_frame();
update_frame(0);
return true;
}
////////////////////////////////////////////////////////////////////
// Function: FFMpegTexture::do_load_one
// Access: Protected, Virtual
// Description: Resets the texture (or the particular level of the
// texture) to the indicated static image.
////////////////////////////////////////////////////////////////////
bool FFMpegTexture::
do_load_one(const PNMImage &pnmimage, const string &name, int z, int n,
const LoaderOptions &options) {
if (z <= (int)_pages.size()) {
VideoPage &page = modify_page(z);
page._color.clear();
}
return Texture::do_load_one(pnmimage, name, z, n, options);
}
////////////////////////////////////////////////////////////////////
// Function: FFMpegTexture::register_with_read_factory
// Access: Public, Static
// Description: Factory method to generate a Texture object
////////////////////////////////////////////////////////////////////
void FFMpegTexture::
register_with_read_factory() {
// Since Texture is such a funny object that is reloaded from the
// TexturePool each time, instead of actually being read fully from
// the bam file, and since the VideoTexture and FFMpegTexture
// classes don't really add any useful data to the bam record, we
// don't need to define make_from_bam(), fillin(), or
// write_datagram() in this class--we just inherit the same
// functions from Texture.
// We do, however, have to register this class with the BamReader,
// to avoid warnings about creating the wrong kind of object from
// the bam file.
BamReader::get_factory()->register_factory(get_class_type(), make_from_bam);
}
////////////////////////////////////////////////////////////////////
// Function: FFMpegTexture::VideoStream::Constructor
// Access: Public
// Description:
////////////////////////////////////////////////////////////////////
FFMpegTexture::VideoStream::
VideoStream() :
_codec_context(NULL),
_format_context(NULL),
_frame(NULL),
_frame_out(NULL),
_next_frame_number(0)
{
// printf("creating video stream\n");
}
////////////////////////////////////////////////////////////////////
// Function: FFMpegTexture::VideoStream::Copy Constructor
// Access: Public
// Description:
////////////////////////////////////////////////////////////////////
FFMpegTexture::VideoStream::
VideoStream(const FFMpegTexture::VideoStream &copy) :
_codec_context(NULL),
_format_context(NULL),
_frame(NULL),
_frame_out(NULL),
_next_frame_number(0)
{
// Rather than copying the _capture pointer, we must open a new
// stream that references the same file.
if (copy.is_valid()) {
if (copy.is_from_file()) {
read(copy._filename);
}
}
}
////////////////////////////////////////////////////////////////////
// Function: FFMpegTexture::VideoStream::Copy Constructor
// Access: Public
// Description:
////////////////////////////////////////////////////////////////////
FFMpegTexture::VideoStream::
~VideoStream() {
clear();
}
////////////////////////////////////////////////////////////////////
// Function: FFMpegTexture::VideoStream::get_frame_data
// Access: Public
// Description: Returns the pointer to the beginning of the
// decompressed buffer for the indicated frame number.
// It is most efficient to call this in increasing order
// of frame number.
////////////////////////////////////////////////////////////////////
bool FFMpegTexture::VideoStream::
get_frame_data(int frame_number) {
nassertr(is_valid(), false);
int coming_from = _next_frame_number;
_next_frame_number = frame_number + 1;
AVPacket packet;
AVStream *vstream = _format_context->streams[_stream_number];
int got_frame;
// Can we get to our target frame just by skipping forward a few
// frames? We arbitrarily draw the line at 50 frames for now.
if (frame_number >= coming_from && frame_number - coming_from < 50) {
if (frame_number > coming_from) {
// Ok, we do have to skip a few frames.
_codec_context->hurry_up = true;
while (frame_number > coming_from) {
int err = read_video_frame(&packet);
if (err < 0) {
return false;
}
avcodec_decode_video(_codec_context, _frame, &got_frame, packet.data,
packet.size);
av_free_packet(&packet);
++coming_from;
}
_codec_context->hurry_up = false;
}
// Now we're ready to read a frame.
int err = read_video_frame(&packet);
if (err < 0) {
return false;
}
} else {
// We have to skip backward, or maybe forward a whole bunch of
// frames. Better off seeking through the stream.
double time_stamp = ((double)AV_TIME_BASE * frame_number * vstream->r_frame_rate.den) / vstream->r_frame_rate.num;
double curr_time_stamp;
// find point in time
av_seek_frame(_format_context, -1, (long long)time_stamp,
AVSEEK_FLAG_BACKWARD);
// Okay, now we're at the nearest keyframe behind our timestamp.
// Hurry up and move through frames until we find a frame just after it.
_codec_context->hurry_up = true;
do {
int err = read_video_frame(&packet);
if (err < 0) {
return false;
}
curr_time_stamp = (((double)AV_TIME_BASE * packet.pts) /
((double)packet.duration * av_q2d(vstream->r_frame_rate)));
if (curr_time_stamp > time_stamp) {
break;
}
avcodec_decode_video(_codec_context, _frame, &got_frame, packet.data,
packet.size);
av_free_packet(&packet);
} while (true);
_codec_context->hurry_up = false;
// Now near frame with Packet ready for decode (and free)
}
// Now we have a packet from someone. Lets get this in a frame
int frame_finished;
// Is this a packet from the video stream?
if (packet.stream_index == _stream_number) {
// Decode video frame
avcodec_decode_video(_codec_context, _frame, &frame_finished,
packet.data, packet.size);
// Did we get a video frame?
if (frame_finished) {
// Convert the image from its native format to RGB
if (_codec_context->pix_fmt != PIX_FMT_RGBA32) {
img_convert((AVPicture *)_frame_out, PIX_FMT_BGR24,
(AVPicture *)_frame, _codec_context->pix_fmt,
_codec_context->width, _codec_context->height);
} else { // _codec_context->pix_fmt == PIX_FMT_RGBA32
img_convert((AVPicture *)_frame_out, PIX_FMT_RGBA32,
(AVPicture *)_frame, _codec_context->pix_fmt,
_codec_context->width, _codec_context->height);
}
}
}
// Free the packet that was allocated by av_read_frame
av_free_packet(&packet);
return true;
}
////////////////////////////////////////////////////////////////////
// Function: FFMpegTexture::VideoStream::read
// Access: Public
// Description: Sets up the stream to read the indicated file.
// Returns true on success, false on failure.
////////////////////////////////////////////////////////////////////
bool FFMpegTexture::VideoStream::
read(const Filename &filename) {
// Clear out the last stream
clear();
string os_specific = filename.to_os_specific();
// Open video file
if (av_open_input_file(&_format_context, os_specific.c_str(), NULL,
0, NULL) != 0) {
// Don't do anything, because nothing happened yet
return false;
}
// Retrieve stream information
if (av_find_stream_info(_format_context) < 0) {
clear();
return false;
}
dump_format(_format_context, 0, os_specific.c_str(), false);
_stream_number = -1;
for(int i = 0; i < _format_context->nb_streams; i++) {
if ((*_format_context->streams[i]->codec).codec_type == CODEC_TYPE_VIDEO) {
_stream_number = i;
break;
}
}
if (_stream_number == -1) {
clear();
return false;
}
// Get a pointer to the codec context for the video stream
AVCodecContext *codec_context = _format_context->streams[_stream_number]->codec;
// Find the decoder for the video stream
// printf("codec id is %d\n",codec_context->codec_id);
_codec = avcodec_find_decoder(codec_context->codec_id);
if (_codec == NULL) {
clear();
return false;
}
if (_codec->capabilities & CODEC_CAP_TRUNCATED) {
codec_context->flags |= CODEC_FLAG_TRUNCATED;
}
// Open codec
_codec_context = codec_context;
if (avcodec_open(_codec_context, _codec) < 0) {
_codec_context = NULL;
clear();
return false;
}
_frame = avcodec_alloc_frame();
if (_codec_context->pix_fmt != PIX_FMT_RGBA32) {
_frame_out = avcodec_alloc_frame();
if (_frame_out == NULL) {
clear();
return false;
}
// Determine required buffer size and allocate buffer
_image_size_bytes = avpicture_get_size(PIX_FMT_BGR24, _codec_context->width,
_codec_context->height);
_raw_data = new uint8_t[_image_size_bytes];
// Assign appropriate parts of buffer to image planes in _frameRGB
avpicture_fill((AVPicture *)_frame_out, _raw_data, PIX_FMT_BGR24,
_codec_context->width, _codec_context->height);
} else {
_frame_out = avcodec_alloc_frame();
if (_frame_out == NULL) {
clear();
return false;
}
// Determine required buffer size and allocate buffer
_image_size_bytes = avpicture_get_size(PIX_FMT_RGBA32, _codec_context->width,
_codec_context->height);
_raw_data = new uint8_t[_image_size_bytes];
// Assign appropriate parts of buffer to image planes in _frameRGB
avpicture_fill((AVPicture *)_frame_out, _raw_data, PIX_FMT_RGBA32,
_codec_context->width, _codec_context->height);
}
// We could put an option here for single channel frames.
_next_frame_number = 0;
_filename = filename;
return true;
}
////////////////////////////////////////////////////////////////////
// Function: FFMpegTexture::VideoStream::clear
// Access: Public
// Description: Stops the video playback and frees the associated
// resources.
////////////////////////////////////////////////////////////////////
void FFMpegTexture::VideoStream::
clear() {
if (_codec_context) {
avcodec_close(_codec_context);
_codec_context = NULL;
}
if (_format_context) {
av_close_input_file(_format_context);
_format_context = NULL;
}
if (_frame) {
av_free(_frame);
_frame = NULL;
}
if (_frame_out) {
av_free(_frame_out);
_frame_out = NULL;
}
_next_frame_number = 0;
}
////////////////////////////////////////////////////////////////////
// Function: FFMpegTexture::VideoStream::read_video_frame
// Access: Private
// Description: Fills packet with the next sequential video frame in
// the stream, skipping over all non-video frames.
// packet must later be deallocated with
// av_free_packet().
//
// Returns nonnegative on success, or negative on error.
////////////////////////////////////////////////////////////////////
int FFMpegTexture::VideoStream::
read_video_frame(AVPacket *packet) {
int err = av_read_frame(_format_context, packet);
if (err < 0) {
return err;
}
while (packet->stream_index != _stream_number) {
// It's not a video packet; free it and get another.
av_free_packet(packet);
err = av_read_frame(_format_context, packet);
if (err < 0) {
grutil_cat.debug()
<< "Got error " << err << " reading frame.\n";
return err;
}
}
// This is a video packet, return it.
return err;
}
#endif // HAVE_FFMpeg