FromSim2Real / gpudrive-main /src /json_serialization.hpp
lzhts1's picture
Upload 385 files
9897e20 verified
Raw
History Blame Contribute Delete
15.7 kB
#pragma once
#include "init.hpp"
#include "types.hpp"
#include "consts.hpp"
#include <iostream>
#include <nlohmann/json.hpp>
#include <unordered_set>
namespace madrona_gpudrive
{
void from_json(const nlohmann::json &j, MapVector2 &p)
{
p.x = j.at("x").get<float>();
p.y = j.at("y").get<float>();
}
void from_json(const nlohmann::json &j, MapObject &obj)
{
obj.mean = {0,0};
uint32_t i = 0;
for (const auto &pos : j.at("position"))
{
if (i < MAX_POSITIONS)
{
from_json(pos, obj.position[i]);
obj.mean.x += (obj.position[i].x - obj.mean.x)/(i+1);
obj.mean.y += (obj.position[i].y - obj.mean.y)/(i+1);
++i;
}
else
{
break; // Avoid overflow
}
}
obj.numPositions = i;
j.at("width").get_to(obj.vehicle_size.width);
j.at("length").get_to(obj.vehicle_size.length);
j.at("height").get_to(obj.vehicle_size.height);
j.at("id").get_to(obj.id);
i = 0;
for (const auto &h : j.at("heading"))
{
if (i < MAX_POSITIONS)
{
h.get_to(obj.heading[i]);
++i;
}
else
{
break; // Avoid overflow
}
}
obj.numHeadings = i;
i = 0;
for (const auto &v : j.at("velocity"))
{
if (i < MAX_POSITIONS)
{
from_json(v, obj.velocity[i]);
++i;
}
else
{
break; // Avoid overflow
}
}
obj.numVelocities = i;
i = 0;
for (const auto &v : j.at("valid"))
{
if (i < MAX_POSITIONS)
{
v.get_to(obj.valid[i]);
++i;
}
else
{
break; // Avoid overflow
}
}
obj.numValid = i;
from_json(j.at("goalPosition"), obj.goalPosition);
std::string type = j.at("type");
if(type == "vehicle")
obj.type = EntityType::Vehicle;
else if(type == "pedestrian")
obj.type = EntityType::Pedestrian;
else if(type == "cyclist")
obj.type = EntityType::Cyclist;
else
obj.type = EntityType::None;
std::string markAsExpertKey = "mark_as_expert";
if (j.contains(markAsExpertKey)) {
from_json(j.at("mark_as_expert"), obj.markAsExpert);
}
// Initialize metadata fields to 0
obj.metadata.isSdc = 0;
obj.metadata.isObjectOfInterest = 0;
obj.metadata.isTrackToPredict = 0;
obj.metadata.difficulty = 0;
}
void from_json(const nlohmann::json &j, MapRoad &road, float polylineReductionThreshold = 0.0)
{
road.mean = {0,0};
std::string type = j.at("type");
if(type == "road_edge")
road.type = EntityType::RoadEdge;
else if(type == "road_line")
road.type = EntityType::RoadLine;
else if(type == "lane")
road.type = EntityType::RoadLane;
else if(type == "crosswalk")
road.type = EntityType::CrossWalk;
else if(type == "speed_bump")
road.type = EntityType::SpeedBump;
else if(type == "stop_sign")
road.type = EntityType::StopSign;
else
road.type = EntityType::None;
std::vector<MapVector2> geometry_points_;
for(const auto &point: j.at("geometry"))
{
MapVector2 p;
from_json(point, p);
geometry_points_.push_back(p);
}
const int64_t num_segments = j["geometry"].size() - 1;
const int64_t sample_every_n_ = 1;
const int64_t num_sampled_points = (num_segments + sample_every_n_ - 1) / sample_every_n_ + 1;
if (num_segments >= 10 && (road.type == EntityType::RoadLane || road.type == EntityType::RoadEdge || road.type == EntityType::RoadLine))
{
std::vector<bool> skip(num_sampled_points, false); // This list tracks the points that are skipped
int64_t k = 0;
bool skipChanged = true; // This is used to check if the skip list has changed in the last iteration
while (skipChanged) // This loop runs O(N^2) in worst case, but it is very fast in practice probably O(NlogN)
{
skipChanged = false; // Reset the skipChanged flag
k = 0;
while (k < num_sampled_points - 1)
{
int64_t k_1 = k + 1; // k_1 is the next point that is not skipped
while (k_1 < num_sampled_points - 1 && skip[k_1])
{
k_1++; // Keep incrementing k_1 until we find a point that is not skipped
}
if (k_1 >= num_sampled_points - 1)
break;
int64_t k_2 = k_1 + 1;
while (k_2 < num_sampled_points && skip[k_2])
{
k_2++; // Keep incrementing k_2 until we find a point that is not skipped
}
if (k_2 >= num_sampled_points)
break;
auto point1 = geometry_points_[k * sample_every_n_];
auto point2 = geometry_points_[k_1 * sample_every_n_];
auto point3 = geometry_points_[k_2 * sample_every_n_];
float_t area = 0.5 * std::abs((point1.x - point3.x) * (point2.y - point1.y) - (point1.x - point2.x) * (point3.y - point1.y));
if (area < polylineReductionThreshold)
{ // If the area is less than the threshold, then we skip the middle point
skip[k_1] = true; // Mark the middle point as skipped
k = k_2; // Skip the middle point and start from the next point
skipChanged = true; // Set the skipChanged flag to true
}
else
{
k = k_1; // If the area is greater than the threshold, then we don't skip the middle point and start from the next point
}
}
}
// Create the road lines
k = 0;
skip[0] = false;
skip[num_sampled_points - 1] = false;
std::vector<MapVector2> new_geometry_points; // This list stores the points that are not skipped
while (k < num_sampled_points)
{
if (!skip[k])
{
new_geometry_points.push_back(geometry_points_[k * sample_every_n_]); // Add the point to the list if it is not skipped
}
k++;
}
for (size_t i = 0; i < new_geometry_points.size(); i++)
{
if(i==MAX_GEOMETRY)
break;
road.geometry[i] = new_geometry_points[i]; // Create the road lines
}
road.numPoints = new_geometry_points.size();
}
else
{
for (int64_t i = 0; i < num_sampled_points ; ++i)
{
if(i==MAX_GEOMETRY)
break;
road.geometry[i] = geometry_points_[i * sample_every_n_];
}
road.numPoints = num_sampled_points;
}
if (j.contains("id")) {
road.id = j.at("id").get<uint32_t>();
}
if (j.contains("map_element_id"))
{
auto mapElementId = j.at("map_element_id").get<int32_t>();
if(mapElementId == 4 or mapElementId >= static_cast<int32_t>(MapType::NUM_TYPES) or mapElementId < -1)
{
road.mapType = MapType::UNKNOWN;
}
else
{
road.mapType = static_cast<MapType>(mapElementId);
}
}
else
{
road.mapType = MapType::UNKNOWN;
}
for (int i = 0; i < road.numPoints; i++)
{
road.mean.x += (road.geometry[i].x - road.mean.x)/(i+1);
road.mean.y += (road.geometry[i].y - road.mean.y)/(i+1);
}
}
std::pair<float, float> calc_mean(const nlohmann::json &j)
{
std::pair<float, float> mean = {0, 0};
int64_t numEntities = 0;
for (const auto &obj : j["objects"])
{
int i = 0;
for (const auto &pos : obj["position"])
{
if(obj["valid"][i++] == false)
continue;
numEntities++;
float newX = pos["x"];
float newY = pos["y"];
// Update mean incrementally
mean.first += (newX - mean.first) / numEntities;
mean.second += (newY - mean.second) / numEntities;
}
}
for (const auto &obj : j["roads"])
{
for (const auto &point : obj["geometry"])
{
numEntities++;
float newX = point["x"];
float newY = point["y"];
// Update mean incrementally
mean.first += (newX - mean.first) / numEntities;
mean.second += (newY - mean.second) / numEntities;
}
}
return mean;
}
void from_json(const nlohmann::json &j, Map &map, float polylineReductionThreshold)
{
std::string name = j.at("name").get<std::string>();
std::strncpy(map.mapName, name.c_str(), sizeof(map.mapName));
std::string scenario_id = j.at("scenario_id").get<std::string>();
std::strncpy(map.scenarioId, scenario_id.c_str(), sizeof(map.scenarioId));
auto mean = calc_mean(j);
map.mean = {mean.first, mean.second};
map.numObjects = std::min(j.at("objects").size(), static_cast<size_t>(MAX_OBJECTS));
const auto& metadata = j.at("metadata");
int sdc_index = metadata.at("sdc_track_index").get<int>();
// Create id to object index mapping
std::unordered_map<int, size_t> idToObjIdx;
size_t idx = 0;
// First, identify which objects are tracks_to_predict and objects_of_interest
std::unordered_set<int> tracks_to_predict_indices;
std::unordered_set<int> objects_of_interest_ids;
// Collect tracks_to_predict indices
for (const auto& track : metadata.at("tracks_to_predict")) {
int track_index = track.at("track_index").get<int>();
if (track_index >= 0 && track_index < j.at("objects").size()) {
tracks_to_predict_indices.insert(track_index);
} else {
std::cerr << "Warning: Invalid track_index " << track_index << " in scene " << j.at("name").get<std::string>() << std::endl;
}
}
// Collect objects_of_interest IDs
for (const auto& obj_id : metadata.at("objects_of_interest")) {
objects_of_interest_ids.insert(obj_id.get<int>());
}
// Initialize SDC first if valid
if (sdc_index >= 0 && sdc_index < j.at("objects").size()) {
j.at("objects")[sdc_index].get_to(map.objects[0]);
map.objects[0].metadata.isSdc = 1;
// Set additional metadata if needed
int sdc_id = map.objects[0].id;
if (tracks_to_predict_indices.find(sdc_index) != tracks_to_predict_indices.end()) {
map.objects[0].metadata.isTrackToPredict = 1;
// Find and set difficulty
for (const auto& track : metadata.at("tracks_to_predict")) {
if (track.at("track_index").get<int>() == sdc_index) {
map.objects[0].metadata.difficulty = track.at("difficulty").get<int>();
break;
}
}
}
if (objects_of_interest_ids.find(sdc_id) != objects_of_interest_ids.end()) {
map.objects[0].metadata.isObjectOfInterest = 1;
}
idToObjIdx[sdc_id] = 0;
idx = 1;
// Remove SDC from sets to avoid double processing
tracks_to_predict_indices.erase(sdc_index);
objects_of_interest_ids.erase(sdc_id);
}
// Initialize tracks_to_predict objects (excluding SDC)
for (size_t i = 0; i < j.at("objects").size() && idx < map.numObjects; i++) {
if (i == sdc_index) continue; // Skip SDC as it's already initialized
if (tracks_to_predict_indices.find(i) != tracks_to_predict_indices.end()) {
j.at("objects")[i].get_to(map.objects[idx]);
map.objects[idx].metadata.isTrackToPredict = 1;
// Find and set difficulty
for (const auto& track : metadata.at("tracks_to_predict")) {
if (track.at("track_index").get<int>() == static_cast<int>(i)) {
map.objects[idx].metadata.difficulty = track.at("difficulty").get<int>();
break;
}
}
// Check if also object of interest
if (objects_of_interest_ids.find(map.objects[idx].id) != objects_of_interest_ids.end()) {
map.objects[idx].metadata.isObjectOfInterest = 1;
objects_of_interest_ids.erase(map.objects[idx].id);
}
idToObjIdx[map.objects[idx].id] = idx;
idx++;
}
}
// Initialize objects_of_interest (excluding those already processed)
for (size_t i = 0; i < j.at("objects").size() && idx < map.numObjects; i++) {
if (i == sdc_index) continue;
int obj_id = j.at("objects")[i].at("id").get<int>();
if (objects_of_interest_ids.find(obj_id) != objects_of_interest_ids.end()) {
j.at("objects")[i].get_to(map.objects[idx]);
map.objects[idx].metadata.isObjectOfInterest = 1;
idToObjIdx[map.objects[idx].id] = idx;
idx++;
}
}
// Initialize all remaining objects
for (size_t i = 0; i < j.at("objects").size() && idx < map.numObjects; i++) {
if (i == sdc_index) continue;
int obj_id = j.at("objects")[i].at("id").get<int>();
if (idToObjIdx.find(obj_id) == idToObjIdx.end()) { // Check if not already processed
j.at("objects")[i].get_to(map.objects[idx]);
idToObjIdx[map.objects[idx].id] = idx;
idx++;
}
}
// Process roads
map.numRoads = std::min(j.at("roads").size(), static_cast<size_t>(MAX_ROADS));
size_t countRoadPoints = 0;
idx = 0;
for (const auto &road : j.at("roads")) {
if (idx >= map.numRoads)
break;
from_json(road, map.roads[idx], polylineReductionThreshold);
size_t roadPoints = map.roads[idx].numPoints;
countRoadPoints += (map.roads[idx].type <= EntityType::RoadLane) ? (roadPoints - 1) : 1;
++idx;
}
map.numRoadSegments = countRoadPoints;
}
}