fix(chain): empty predictions were too empty

src/backends/tensorrt/tensorrtlib.cc

@@ -346,10 +346,13 @@ namespace dd
                    TMLModel>::clear_mllib(const APIData &ad)
   {
     (void)ad;
-    cudaFree(_buffers.data()[_inputIndex]);
-    cudaFree(_buffers.data()[_outputIndex0]);
-    if (_bbox)
-      cudaFree(_buffers.data()[_outputIndex1]);
+    if (!_buffers.empty())
+      {
+        cudaFree(_buffers.at(_inputIndex));
+        cudaFree(_buffers.at(_outputIndex0));
+        if (_bbox)
+          cudaFree(_buffers.at(_outputIndex1));
+      }
 
     // remove compiled model files.
     std::vector<std::string> extensions

src/services.h

@@ -842,9 +842,8 @@ namespace dd
         {
           chain_logger->info("[" + std::to_string(chain_pos)
                              + "] / no result from prediction");
-          cdata.add_model_data(
-              pred_id,
-              DTO::PredictBody::createShared()); // store empty model output
+          cdata.add_model_data(pred_id,
+                               pred_dto); // store empty model output
           return 1;
         }
 
@@ -1170,9 +1169,8 @@ namespace dd
         {
           chain_logger->info("[" + std::to_string(chain_pos)
                              + "] / no result from prediction");
-          cdata.add_model_data(
-              pred_id,
-              DTO::PredictBody::createShared()); // store empty model output
+          cdata.add_model_data(pred_id,
+                               pred_dto); // store empty model output
           return 1;
         }
 

tests/ut-chain.cc

@@ -127,6 +127,30 @@ TEST(chain, chain_torch_detection_classification)
                 .GetString(),
             std::string("n02086079 Pekinese, Pekingese, Peke"));
 
+  // chain predict without detection
+  jchainstr
+      = "{\"chain\":{\"name\":\"chain\",\"calls\":["
+        "{\"service\":\""
+        + detect_sname
+        + "\",\"parameters\":{\"input\":{\"keep_orig\":true},\"output\":{"
+          "\"bbox\":true,\"confidence_threshold\":0.9999}},\"data\":[\""
+        + uri1
+        + "\"]},"
+          "{\"id\":\"crop\",\"action\":{\"type\":\"crop\",\"parameters\":{"
+          "\"padding_ratio\":0.05}}},{\"service\":\""
+        + classif_sname
+        + "\",\"parent_id\":\"crop\",\"parameters\":{\"output\":{\"best\":1}}}"
+          "]}}";
+  joutstr = japi.jrender(japi.service_chain("chain", jchainstr));
+  jd = JDoc();
+  std::cout << "joutstr=" << joutstr << std::endl;
+  jd.Parse<rapidjson::kParseNanAndInfFlag>(joutstr.c_str());
+  ASSERT_TRUE(!jd.HasParseError());
+  ASSERT_EQ(200, jd["status"]["code"]);
+  ASSERT_TRUE(jd["body"]["predictions"].IsArray());
+  ASSERT_EQ(jd["body"]["predictions"].Size(), 1);
+  ASSERT_EQ(jd["body"]["predictions"][0]["classes"].Size(), 0);
+
   // multiple models (tree)
   std::string classif2_sname = "classif2";
   jstr = "{\"mllib\":\"torch\",\"description\":\"squeezenet\",\"type\":"
@@ -181,7 +205,6 @@ TEST(chain, chain_torch_detection_classification)
   // cleanup
   fileops::remove_file(torch_detect_repo, "model.json");
 }
-
 #endif
 
 #ifdef USE_CAFFE
@@ -486,4 +509,82 @@ TEST(chain, chain_trt_detection_gan)
                                     + get_trt_archi() + "_bs1"));
 }
 
+// Test internal call without json
+TEST(chain, chain_trt_dto)
+{
+  JsonAPI japi;
+  std::string detect_sname = "detect";
+  std::string jstr
+      = "{\"mllib\":\"tensorrt\",\"description\":\"yolox\","
+        "\"type\":\"supervised\",\"model\":{\"repository\":\""
+        + trt_detect_repo
+        + "\"},\"parameters\":{\"input\":{\"connector\":"
+          "\"image\",\"height\":640,\"width\":640},\"mllib\":{"
+          "\"maxWorkspaceSize\":256,\"gpuid\":0,"
+          "\"template\":\"yolox\",\"nclasses\":81,\"datatype\":\"fp16\"}}}";
+  std::string joutstr = japi.jrender(japi.service_create(detect_sname, jstr));
+  ASSERT_EQ(created_str, joutstr);
+
+  std::string gan_sname = "gan";
+  jstr = "{\"mllib\":\"tensorrt\",\"description\":\"gan\",\"type\":"
+         "\"supervised\",\"model\":{\"repository\":\""
+         + trt_gan_repo
+         + "\"},\"parameters\":{\"input\":{\"connector\":\"image\",\"height\":"
+           "360,\"width\":360,\"rgb\":true,\"scale\":0.0039,\"mean\":[0.5, "
+           "0.5,0.5],\"std\":[0.5,0.5,0.5]},\"mllib\":{\"maxBatchSize\":1,"
+           "\"maxWorkspaceSize\":256,\"gpuid\":0,\"datatype\":\"fp16\"}}}";
+  joutstr = japi.jrender(japi.service_create(gan_sname, jstr));
+  ASSERT_EQ(created_str, joutstr);
+
+  // chain call with no predictions
+  std::string uri1 = trt_gan_repo + "/horse_1024.jpg";
+  auto input_dto = oatpp::Object<DTO::ServiceChain>::createShared();
+  input_dto->chain = oatpp::Object<DTO::Chain>::createShared();
+
+  auto call1 = oatpp::Object<DTO::ChainCall>::createShared();
+  call1->service = detect_sname;
+  call1->parameters->input->keep_orig = true;
+  call1->parameters->output->bbox = true;
+  call1->parameters->output->confidence_threshold = 0.9999;
+  call1->data->push_back(uri1);
+  input_dto->chain->calls->push_back(call1);
+
+  auto call2 = oatpp::Object<DTO::ChainCall>::createShared();
+  call2->id = "crop";
+  call2->action = oatpp::Object<DTO::ChainAction>::createShared();
+  call2->action->type = "crop";
+  call2->action->parameters->padding_ratio = 0.05;
+  input_dto->chain->calls->push_back(call2);
+
+  auto call3 = oatpp::Object<DTO::ChainCall>::createShared();
+  call3->service = gan_sname;
+  call3->parent_id = "crop";
+  call3->parameters->mllib->extract_layer = "last";
+  call3->parameters->output->image = true;
+  input_dto->chain->calls->push_back(call3);
+
+  auto chain_out = japi.chain(input_dto, "chain");
+  JDoc jdoc;
+  oatpp_utils::dtoToJDoc(chain_out, jdoc);
+  std::cout << dd_utils::jrender(jdoc) << std::endl;
+
+  ASSERT_EQ(chain_out->predictions->size(), 1);
+  // PredictClass -> only one model, so the Predict DTO is returned without
+  // modifications
+  ASSERT_EQ((*chain_out->predictions->at(0))["classes"]
+                .retrieve<oatpp::Vector<oatpp::Object<DTO::PredictClass>>>()
+                ->size(),
+            0);
+
+  jstr = "{\"clear\":\"lib\"}";
+  joutstr = japi.jrender(japi.service_delete(detect_sname, jstr));
+  ASSERT_EQ(ok_str, joutstr);
+  joutstr = japi.jrender(japi.service_delete(gan_sname, jstr));
+  ASSERT_EQ(ok_str, joutstr);
+  ASSERT_TRUE(!fileops::file_exists(trt_detect_repo + "/TRTengine_arch"
+                                    + get_trt_archi() + "_bs1"));
+  ASSERT_TRUE(!fileops::file_exists(trt_gan_repo + "/TRTengine_arch"
+                                    + get_trt_archi() + "_bs1"));
+}
+
 #endif