Fix GT jobs creation (#7126)

- Fixed gt job creation for the whole task size
- Fixed invalid chunk writing for GT jobs

changelog.d/20231110_183941_mzhiltsov_fix_gt_jobs.md

@@ -0,0 +1,6 @@
+### Fixed
+
+- It is now possible to create Ground Truth jobs containing all frames in the task
+  (<https://github.com/opencv/cvat/pull/7126>)
+- Incorrect Ground Truth chunks saving
+  (<https://github.com/opencv/cvat/pull/7126>)

cvat/apps/engine/cache.py

@@ -65,7 +65,7 @@ def get_task_chunk_data_with_mime(self, chunk_number, quality, db_data):
 
     def get_selective_job_chunk_data_with_mime(self, chunk_number, quality, job):
         item = self._get_or_set_cache_item(
-            key=f'{job.id}_{chunk_number}_{quality}',
+            key=f'job_{job.id}_{chunk_number}_{quality}',
             create_function=lambda: self.prepare_selective_job_chunk(job, quality, chunk_number),
         )
 

cvat/apps/engine/models.py

@@ -242,7 +242,7 @@ def get_frame_step(self):
         return int(match.group(1)) if match else 1
 
     def get_valid_frame_indices(self):
-        return range(self.start_frame, self.stop_frame, self.get_frame_step())
+        return range(self.start_frame, self.stop_frame + 1, self.get_frame_step())
 
     def get_data_dirname(self):
         return os.path.join(settings.MEDIA_DATA_ROOT, str(self.id))
@@ -599,7 +599,7 @@ def clean(self) -> None:
             )
 
         if self.stop_frame < self.start_frame:
-            raise ValidationError("stop_frame cannot be lesser than start_frame")
+            raise ValidationError("stop_frame cannot be less than start_frame")
 
         return super().clean()
 

cvat/apps/engine/serializers.py

@@ -638,10 +638,10 @@ def create(self, validated_data):
             frame_selection_method = validated_data.pop("frame_selection_method", None)
             if frame_selection_method == models.JobFrameSelectionMethod.RANDOM_UNIFORM:
                 frame_count = validated_data.pop("frame_count")
-                if size <= frame_count:
+                if size < frame_count:
                     raise serializers.ValidationError(
-                        f"The number of frames requested ({frame_count}) must be lesser than "
-                        f"the number of the task frames ({size})"
+                        f"The number of frames requested ({frame_count}) "
+                        f"must be not be greater than the number of the task frames ({size})"
                     )
 
                 seed = validated_data.pop("seed", None)
@@ -650,6 +650,13 @@ def create(self, validated_data):
                 # so here we specify it explicitly
                 from numpy import random
                 rng = random.Generator(random.MT19937(seed=seed))
+
+                if seed is not None and frame_count < size:
+                    # Reproduce the old (a little bit incorrect) behavior that existed before
+                    # https://github.com/opencv/cvat/pull/7126
+                    # to make the old seed-based sequences reproducible
+                    valid_frame_ids = [v for v in valid_frame_ids if v != task.data.stop_frame]
+
                 frames = rng.choice(
                     list(valid_frame_ids), size=frame_count, shuffle=False, replace=False
                 ).tolist()

tests/python/rest_api/test_jobs.py

@@ -171,7 +171,34 @@ def test_can_create_gt_job_with_random_frames_and_seed(self, admin_user, task_id
         with make_api_client(user) as api_client:
             (gt_job_meta, _) = api_client.jobs_api.retrieve_data_meta(job_id)
 
-        assert gt_job_meta.included_frames == frame_ids
+        assert frame_ids == gt_job_meta.included_frames
+
+    @pytest.mark.parametrize("task_mode", ["annotation", "interpolation"])
+    def test_can_create_gt_job_with_all_frames(self, admin_user, tasks, jobs, task_mode):
+        user = admin_user
+        task = next(
+            t
+            for t in tasks
+            if t["mode"] == task_mode
+            and t["size"]
+            and not any(j for j in jobs if j["task_id"] == t["id"] and j["type"] == "ground_truth")
+        )
+        task_id = task["id"]
+
+        job_spec = {
+            "task_id": task_id,
+            "type": "ground_truth",
+            "frame_selection_method": "random_uniform",
+            "frame_count": task["size"],
+        }
+
+        response = self._test_create_job_ok(user, job_spec)
+        job_id = json.loads(response.data)["id"]
+
+        with make_api_client(user) as api_client:
+            (gt_job_meta, _) = api_client.jobs_api.retrieve_data_meta(job_id)
+
+        assert task["size"] == gt_job_meta.size
 
     def test_can_create_no_more_than_1_gt_job(self, admin_user, jobs):
         user = admin_user