sadmen
/
openpilot_comma
mirror of https://github.com/commaai/openpilot.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

ModelDataRaw struct part 4 (#22673)

Browse Source 
* convert model data for leads to struct

* make things more consistent

* change names with path to plan

* consistent use of plural
pull/22694/head
Greg Hogan 4 years ago committed by GitHub
parent 9ca16560e2
commit 7bb13acc45
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
2 changed files with 91 additions and 83 deletions
Whitespace
Show all changes Ignore whitespace when comparing lines Ignore changes in amount of whitespace Ignore changes in whitespace at EOL
Unified View
Diff Options
Show Stats Download Patch File Download Diff File
  1. 114
      selfdrive/modeld/models/driving.cc
  2. 60
      selfdrive/modeld/models/driving.h

114
selfdrive/modeld/models/driving.cc
Unescape View File

@ -77,11 +77,10 @@ ModelDataRaw model_eval_frame(ModelState* s, cl_mem yuv_cl, int width, int heigh
// net outputs // net outputs
ModelDataRaw net_outputs { ModelDataRaw net_outputs {
.plan = (ModelDataRawPlans*)&s->output[PLAN_IDX], .plans = (ModelDataRawPlans*)&s->output[PLAN_IDX],
.lane_lines = (ModelDataRawLaneLines*)&s->output[LL_IDX], .lane_lines = (ModelDataRawLaneLines*)&s->output[LL_IDX],
.road_edges = (ModelDataRawRoadEdges*)&s->output[RE_IDX], .road_edges = (ModelDataRawRoadEdges*)&s->output[RE_IDX],
.lead = &s->output[LEAD_IDX], .leads = (ModelDataRawLeads*)&s->output[LEAD_IDX],
.lead_prob = &s->output[LEAD_PROB_IDX],
.meta = &s->output[DESIRE_STATE_IDX], .meta = &s->output[DESIRE_STATE_IDX],
.pose = (ModelDataRawPose*)&s->output[POSE_IDX], .pose = (ModelDataRawPose*)&s->output[POSE_IDX],
}; };
@ -92,59 +91,38 @@ void model_free(ModelState* s) {
delete s->frame; delete s->frame;
} }
static const float *get_best_data(const float *data, int size, int group_size, int weight_idx) {
int max_idx = 0;
for (int i = 1; i < size; i++) {
if (data[i * group_size + weight_idx] >
data[max_idx * group_size + weight_idx]) {
max_idx = i;
}
}
return &data[max_idx * group_size];
}
static const float *get_lead_data(const float *lead, int t_offset) {
return get_best_data(lead, LEAD_MHP_N, LEAD_MHP_GROUP_SIZE, LEAD_MHP_GROUP_SIZE - LEAD_MHP_SELECTION + t_offset);
}
void fill_sigmoid(const float *input, float *output, int len, int stride) { void fill_sigmoid(const float *input, float *output, int len, int stride) {
for (int i=0; i<len; i++) { for (int i=0; i<len; i++) {
output[i] = sigmoid(input[i*stride]); output[i] = sigmoid(input[i*stride]);
} }
} }
void fill_lead_v3(cereal::ModelDataV2::LeadDataV3::Builder lead, const float *lead_data, const float *prob, int t_offset, float prob_t) { void fill_lead(cereal::ModelDataV2::LeadDataV3::Builder lead, const ModelDataRawLeads &leads, int t_idx, float prob_t) {
float t[LEAD_TRAJ_LEN] = {0.0, 2.0, 4.0, 6.0, 8.0, 10.0}; std::array<float, LEAD_TRAJ_LEN> lead_t = {0.0, 2.0, 4.0, 6.0, 8.0, 10.0};
const float *data = get_lead_data(lead_data, t_offset); auto best_prediction = leads.get_best_prediction(t_idx);
lead.setProb(sigmoid(prob[t_offset])); lead.setProb(sigmoid(leads.prob[t_idx]));
lead.setProbTime(prob_t); lead.setProbTime(prob_t);
float x_arr[LEAD_TRAJ_LEN]; std::array<float, LEAD_TRAJ_LEN> lead_x, lead_y, lead_v, lead_a;
float y_arr[LEAD_TRAJ_LEN]; std::array<float, LEAD_TRAJ_LEN> lead_x_std, lead_y_std, lead_v_std, lead_a_std;
float v_arr[LEAD_TRAJ_LEN];
float a_arr[LEAD_TRAJ_LEN];
float x_stds_arr[LEAD_TRAJ_LEN];
float y_stds_arr[LEAD_TRAJ_LEN];
float v_stds_arr[LEAD_TRAJ_LEN];
float a_stds_arr[LEAD_TRAJ_LEN];
for (int i=0; i<LEAD_TRAJ_LEN; i++) { for (int i=0; i<LEAD_TRAJ_LEN; i++) {
x_arr[i] = data[i*LEAD_PRED_DIM+0]; lead_x[i] = best_prediction.mean[i].x;
y_arr[i] = data[i*LEAD_PRED_DIM+1]; lead_y[i] = best_prediction.mean[i].y;
v_arr[i] = data[i*LEAD_PRED_DIM+2]; lead_v[i] = best_prediction.mean[i].velocity;
a_arr[i] = data[i*LEAD_PRED_DIM+3]; lead_a[i] = best_prediction.mean[i].acceleration;
x_stds_arr[i] = exp(data[LEAD_MHP_VALS + i*LEAD_PRED_DIM+0]); lead_x_std[i] = exp(best_prediction.std[i].x);
y_stds_arr[i] = exp(data[LEAD_MHP_VALS + i*LEAD_PRED_DIM+1]); lead_y_std[i] = exp(best_prediction.std[i].y);
v_stds_arr[i] = exp(data[LEAD_MHP_VALS + i*LEAD_PRED_DIM+2]); lead_v_std[i] = exp(best_prediction.std[i].velocity);
a_stds_arr[i] = exp(data[LEAD_MHP_VALS + i*LEAD_PRED_DIM+3]); lead_a_std[i] = exp(best_prediction.std[i].acceleration);
} }
lead.setT(t); lead.setT(to_kj_array_ptr(lead_t));
lead.setX(x_arr); lead.setX(to_kj_array_ptr(lead_x));
lead.setY(y_arr); lead.setY(to_kj_array_ptr(lead_y));
lead.setV(v_arr); lead.setV(to_kj_array_ptr(lead_v));
lead.setA(a_arr); lead.setA(to_kj_array_ptr(lead_a));
lead.setXStd(x_stds_arr); lead.setXStd(to_kj_array_ptr(lead_x_std));
lead.setYStd(y_stds_arr); lead.setYStd(to_kj_array_ptr(lead_y_std));
lead.setVStd(v_stds_arr); lead.setVStd(to_kj_array_ptr(lead_v_std));
lead.setAStd(a_stds_arr); lead.setAStd(to_kj_array_ptr(lead_a_std));
} }
void fill_meta(cereal::ModelDataV2::MetaData::Builder meta, const float *meta_data) { void fill_meta(cereal::ModelDataV2::MetaData::Builder meta, const float *meta_data) {
@ -219,7 +197,7 @@ void fill_xyzt(cereal::ModelDataV2::XYZTData::Builder xyzt, const std::array<flo
xyzt.setZStd(to_kj_array_ptr(z_std)); xyzt.setZStd(to_kj_array_ptr(z_std));
} }
void fill_plan(cereal::ModelDataV2::Builder &framed, const ModelDataRawPlanPath &plan) { void fill_plan(cereal::ModelDataV2::Builder &framed, const ModelDataRawPlanPrediction &plan) {
std::array<float, TRAJECTORY_SIZE> pos_x, pos_y, pos_z; std::array<float, TRAJECTORY_SIZE> pos_x, pos_y, pos_z;
std::array<float, TRAJECTORY_SIZE> pos_x_std, pos_y_std, pos_z_std; std::array<float, TRAJECTORY_SIZE> pos_x_std, pos_y_std, pos_z_std;
std::array<float, TRAJECTORY_SIZE> vel_x, vel_y, vel_z; std::array<float, TRAJECTORY_SIZE> vel_x, vel_y, vel_z;
@ -250,7 +228,7 @@ void fill_plan(cereal::ModelDataV2::Builder &framed, const ModelDataRawPlanPath
fill_xyzt(framed.initOrientationRate(), T_IDXS_FLOAT, rot_rate_x, rot_rate_y, rot_rate_z); fill_xyzt(framed.initOrientationRate(), T_IDXS_FLOAT, rot_rate_x, rot_rate_y, rot_rate_z);
} }
void fill_lane_lines(cereal::ModelDataV2::Builder &framed, const std::array<float, TRAJECTORY_SIZE> &plan_t_arr, void fill_lane_lines(cereal::ModelDataV2::Builder &framed, const std::array<float, TRAJECTORY_SIZE> &plan_t,
const ModelDataRawLaneLines &lanes) { const ModelDataRawLaneLines &lanes) {
std::array<float, TRAJECTORY_SIZE> left_far_y, left_far_z; std::array<float, TRAJECTORY_SIZE> left_far_y, left_far_z;
std::array<float, TRAJECTORY_SIZE> left_near_y, left_near_z; std::array<float, TRAJECTORY_SIZE> left_near_y, left_near_z;
@ -268,10 +246,10 @@ void fill_lane_lines(cereal::ModelDataV2::Builder &framed, const std::array<floa
} }
auto lane_lines = framed.initLaneLines(4); auto lane_lines = framed.initLaneLines(4);
fill_xyzt(lane_lines[0], plan_t_arr, X_IDXS_FLOAT, left_far_y, left_far_z); fill_xyzt(lane_lines[0], plan_t, X_IDXS_FLOAT, left_far_y, left_far_z);
fill_xyzt(lane_lines[1], plan_t_arr, X_IDXS_FLOAT, left_near_y, left_near_z); fill_xyzt(lane_lines[1], plan_t, X_IDXS_FLOAT, left_near_y, left_near_z);
fill_xyzt(lane_lines[2], plan_t_arr, X_IDXS_FLOAT, right_near_y, right_near_z); fill_xyzt(lane_lines[2], plan_t, X_IDXS_FLOAT, right_near_y, right_near_z);
fill_xyzt(lane_lines[3], plan_t_arr, X_IDXS_FLOAT, right_far_y, right_far_z); fill_xyzt(lane_lines[3], plan_t, X_IDXS_FLOAT, right_far_y, right_far_z);
framed.setLaneLineStds({ framed.setLaneLineStds({
exp(lanes.std.left_far[0].y), exp(lanes.std.left_far[0].y),
@ -288,7 +266,7 @@ void fill_lane_lines(cereal::ModelDataV2::Builder &framed, const std::array<floa
}); });
} }
void fill_road_edges(cereal::ModelDataV2::Builder &framed, const std::array<float, TRAJECTORY_SIZE> &plan_t_arr, void fill_road_edges(cereal::ModelDataV2::Builder &framed, const std::array<float, TRAJECTORY_SIZE> &plan_t,
const ModelDataRawRoadEdges &edges) { const ModelDataRawRoadEdges &edges) {
std::array<float, TRAJECTORY_SIZE> left_y, left_z; std::array<float, TRAJECTORY_SIZE> left_y, left_z;
std::array<float, TRAJECTORY_SIZE> right_y, right_z; std::array<float, TRAJECTORY_SIZE> right_y, right_z;
@ -300,8 +278,8 @@ void fill_road_edges(cereal::ModelDataV2::Builder &framed, const std::array<floa
} }
auto road_edges = framed.initRoadEdges(2); auto road_edges = framed.initRoadEdges(2);
fill_xyzt(road_edges[0], plan_t_arr, X_IDXS_FLOAT, left_y, left_z); fill_xyzt(road_edges[0], plan_t, X_IDXS_FLOAT, left_y, left_z);
fill_xyzt(road_edges[1], plan_t_arr, X_IDXS_FLOAT, right_y, right_z); fill_xyzt(road_edges[1], plan_t, X_IDXS_FLOAT, right_y, right_z);
framed.setRoadEdgeStds({ framed.setRoadEdgeStds({
exp(edges.std.left[0].y), exp(edges.std.left[0].y),
@ -310,18 +288,18 @@ void fill_road_edges(cereal::ModelDataV2::Builder &framed, const std::array<floa
} }
void fill_model(cereal::ModelDataV2::Builder &framed, const ModelDataRaw &net_outputs) { void fill_model(cereal::ModelDataV2::Builder &framed, const ModelDataRaw &net_outputs) {
auto best_plan = net_outputs.plan->get_best_plan(); auto best_plan = net_outputs.plans->get_best_prediction();
std::array<float, TRAJECTORY_SIZE> plan_t_arr; std::array<float, TRAJECTORY_SIZE> plan_t;
std::fill_n(plan_t_arr.data(), plan_t_arr.size(), NAN); std::fill_n(plan_t.data(), plan_t.size(), NAN);
plan_t_arr[0] = 0.0; plan_t[0] = 0.0;
for (int xidx=1, tidx=0; xidx<TRAJECTORY_SIZE; xidx++) { for (int xidx=1, tidx=0; xidx<TRAJECTORY_SIZE; xidx++) {
// increment tidx until we find an element that's further away than the current xidx // increment tidx until we find an element that's further away than the current xidx
for (int next_tid = tidx + 1; next_tid < TRAJECTORY_SIZE && best_plan.mean[next_tid].position.x < X_IDXS[xidx]; next_tid++) { for (int next_tid = tidx + 1; next_tid < TRAJECTORY_SIZE && best_plan.mean[next_tid].position.x < X_IDXS[xidx]; next_tid++) {
tidx++; tidx++;
} }
if (tidx == TRAJECTORY_SIZE - 1) { if (tidx == TRAJECTORY_SIZE - 1) {
// if the plan doesn't extend far enough, set plan_t to the max value (10s), then break // if the Plan doesn't extend far enough, set plan_t to the max value (10s), then break
plan_t_arr[xidx] = T_IDXS[TRAJECTORY_SIZE - 1]; plan_t[xidx] = T_IDXS[TRAJECTORY_SIZE - 1];
break; break;
} }
@ -329,21 +307,21 @@ void fill_model(cereal::ModelDataV2::Builder &framed, const ModelDataRaw &net_ou
float current_x_val = best_plan.mean[tidx].position.x; float current_x_val = best_plan.mean[tidx].position.x;
float next_x_val = best_plan.mean[tidx+1].position.x; float next_x_val = best_plan.mean[tidx+1].position.x;
float p = (X_IDXS[xidx] - current_x_val) / (next_x_val - current_x_val); float p = (X_IDXS[xidx] - current_x_val) / (next_x_val - current_x_val);
plan_t_arr[xidx] = p * T_IDXS[tidx+1] + (1 - p) * T_IDXS[tidx]; plan_t[xidx] = p * T_IDXS[tidx+1] + (1 - p) * T_IDXS[tidx];
} }
fill_plan(framed, best_plan); fill_plan(framed, best_plan);
fill_lane_lines(framed, plan_t_arr, *net_outputs.lane_lines); fill_lane_lines(framed, plan_t, *net_outputs.lane_lines);
fill_road_edges(framed, plan_t_arr, *net_outputs.road_edges); fill_road_edges(framed, plan_t, *net_outputs.road_edges);
// meta // meta
fill_meta(framed.initMeta(), net_outputs.meta); fill_meta(framed.initMeta(), net_outputs.meta);
// leads // leads
auto leads = framed.initLeadsV3(LEAD_MHP_SELECTION); auto leads = framed.initLeadsV3(LEAD_MHP_SELECTION);
float t_offsets[LEAD_MHP_SELECTION] = {0.0, 2.0, 4.0}; std::array<float, LEAD_MHP_SELECTION> t_offsets = {0.0, 2.0, 4.0};
for (int t_offset=0; t_offset<LEAD_MHP_SELECTION; t_offset++) { for (int i=0; i<LEAD_MHP_SELECTION; i++) {
fill_lead_v3(leads[t_offset], net_outputs.lead, net_outputs.lead_prob, t_offset, t_offsets[t_offset]); fill_lead(leads[i], *net_outputs.leads, i, t_offsets[i]);
} }
} }

60
selfdrive/modeld/models/driving.h
Unescape View File

@ -68,36 +68,36 @@ struct ModelDataRawYZ {
}; };
static_assert(sizeof(ModelDataRawYZ) == sizeof(float)*2); static_assert(sizeof(ModelDataRawYZ) == sizeof(float)*2);
struct ModelDataRawPlanTimeStep { struct ModelDataRawPlanElement {
ModelDataRawXYZ position; ModelDataRawXYZ position;
ModelDataRawXYZ velocity; ModelDataRawXYZ velocity;
ModelDataRawXYZ acceleration; ModelDataRawXYZ acceleration;
ModelDataRawXYZ rotation; ModelDataRawXYZ rotation;
ModelDataRawXYZ rotation_rate; ModelDataRawXYZ rotation_rate;
}; };
static_assert(sizeof(ModelDataRawPlanTimeStep) == sizeof(ModelDataRawXYZ)*5); static_assert(sizeof(ModelDataRawPlanElement) == sizeof(ModelDataRawXYZ)*5);
struct ModelDataRawPlanPath { struct ModelDataRawPlanPrediction {
std::array<ModelDataRawPlanTimeStep, TRAJECTORY_SIZE> mean; std::array<ModelDataRawPlanElement, TRAJECTORY_SIZE> mean;
std::array<ModelDataRawPlanTimeStep, TRAJECTORY_SIZE> std; std::array<ModelDataRawPlanElement, TRAJECTORY_SIZE> std;
float prob; float prob;
}; };
static_assert(sizeof(ModelDataRawPlanPath) == (sizeof(ModelDataRawPlanTimeStep)*TRAJECTORY_SIZE*2) + sizeof(float)); static_assert(sizeof(ModelDataRawPlanPrediction) == (sizeof(ModelDataRawPlanElement)*TRAJECTORY_SIZE*2) + sizeof(float));
struct ModelDataRawPlans { struct ModelDataRawPlans {
std::array<ModelDataRawPlanPath, PLAN_MHP_N> path; std::array<ModelDataRawPlanPrediction, PLAN_MHP_N> prediction;
constexpr const ModelDataRawPlanPath &get_best_plan() const { constexpr const ModelDataRawPlanPrediction &get_best_prediction() const {
int max_idx = 0; int max_idx = 0;
for (int i = 1; i < path.size(); i++) { for (int i = 1; i < prediction.size(); i++) {
if (path[i].prob > path[max_idx].prob) { if (prediction[i].prob > prediction[max_idx].prob) {
max_idx = i; max_idx = i;
} }
} }
return path[max_idx]; return prediction[max_idx];
} }
}; };
static_assert(sizeof(ModelDataRawPlans) == sizeof(ModelDataRawPlanPath)*PLAN_MHP_N); static_assert(sizeof(ModelDataRawPlans) == sizeof(ModelDataRawPlanPrediction)*PLAN_MHP_N);
struct ModelDataRawLinesXY { struct ModelDataRawLinesXY {
std::array<ModelDataRawYZ, TRAJECTORY_SIZE> left_far; std::array<ModelDataRawYZ, TRAJECTORY_SIZE> left_far;
@ -140,6 +140,37 @@ struct ModelDataRawRoadEdges {
}; };
static_assert(sizeof(ModelDataRawRoadEdges) == (sizeof(ModelDataRawEdgessXY)*2)); static_assert(sizeof(ModelDataRawRoadEdges) == (sizeof(ModelDataRawEdgessXY)*2));
struct ModelDataRawLeadElement {
float x;
float y;
float velocity;
float acceleration;
};
static_assert(sizeof(ModelDataRawLeadElement) == sizeof(float)*4);
struct ModelDataRawLeadPrediction {
std::array<ModelDataRawLeadElement, LEAD_TRAJ_LEN> mean;
std::array<ModelDataRawLeadElement, LEAD_TRAJ_LEN> std;
std::array<float, LEAD_MHP_SELECTION> prob;
};
static_assert(sizeof(ModelDataRawLeadPrediction) == (sizeof(ModelDataRawLeadElement)*LEAD_TRAJ_LEN*2) + (sizeof(float)*LEAD_MHP_SELECTION));
struct ModelDataRawLeads {
std::array<ModelDataRawLeadPrediction, LEAD_MHP_N> prediction;
std::array<float, LEAD_MHP_SELECTION> prob;
constexpr const ModelDataRawLeadPrediction &get_best_prediction(int t_idx) const {
int max_idx = 0;
for (int i = 1; i < prediction.size(); i++) {
if (prediction[i].prob[t_idx] > prediction[max_idx].prob[t_idx]) {
max_idx = i;
}
}
return prediction[max_idx];
}
};
static_assert(sizeof(ModelDataRawLeads) == (sizeof(ModelDataRawLeadPrediction)*LEAD_MHP_N) + (sizeof(float)*LEAD_MHP_SELECTION));
struct ModelDataRawPose { struct ModelDataRawPose {
ModelDataRawXYZ velocity_mean; ModelDataRawXYZ velocity_mean;
ModelDataRawXYZ rotation_mean; ModelDataRawXYZ rotation_mean;
@ -149,11 +180,10 @@ struct ModelDataRawPose {
static_assert(sizeof(ModelDataRawPose) == sizeof(ModelDataRawXYZ)*4); static_assert(sizeof(ModelDataRawPose) == sizeof(ModelDataRawXYZ)*4);
struct ModelDataRaw { struct ModelDataRaw {
const ModelDataRawPlans *const plan; const ModelDataRawPlans *const plans;
const ModelDataRawLaneLines *const lane_lines; const ModelDataRawLaneLines *const lane_lines;
const ModelDataRawRoadEdges *const road_edges; const ModelDataRawRoadEdges *const road_edges;
const float *const lead; const ModelDataRawLeads *const leads;
const float *const lead_prob;
const float *const desire_state; const float *const desire_state;
const float *const meta; const float *const meta;
const float *const desire_pred; const float *const desire_pred;

Write Preview
Loading…
Cancel
Save