mirror of
https://github.com/blakeblackshear/frigate.git
synced 2026-02-10 13:15:25 +03:00
Remove yolov8 from dev
This commit is contained in:
Nicolas Mowen
parent
94c8edaff7
commit
13e4cbb31b
@ -105,58 +105,6 @@ detectors:
|
||||
device: pci
|
||||
```
|
||||
|
||||
### Yolov8 On Coral
|
||||
|
||||
It is possible to use the [ultralytics yolov8](https://github.com/ultralytics/ultralytics) pretrained models with the Google Coral processors.
|
||||
|
||||
#### Setup
|
||||
|
||||
You need to download yolov8 model files suitable for the EdgeTPU. Frigate can do this automatically with the `DOWNLOAD_YOLOV8={0 | 1}` environment variable either from the command line
|
||||
|
||||
```bash
|
||||
$ docker run ... -e DOWNLOAD_YOLOV8=1 \
|
||||
...
|
||||
```
|
||||
|
||||
or when using docker compose:
|
||||
|
||||
```yaml
|
||||
services:
|
||||
frigate:
|
||||
---
|
||||
environment:
|
||||
DOWNLOAD_YOLOV8: "1"
|
||||
```
|
||||
|
||||
When this variable is set then frigate will at startup fetch [yolov8.small.models.tar.gz](https://github.com/harakas/models/releases/download/yolov8.1-1.1/yolov8.small.models.tar.gz) and extract it into the `/config/model_cache/yolov8/` directory.
|
||||
|
||||
The following files suitable for the EdgeTPU detector will be available under `/config/model_cache/yolov8/`:
|
||||
|
||||
- `yolov8[ns]_320x320_edgetpu.tflite` -- nano (n) and small (s) sized models that have been trained using the coco dataset (90 classes)
|
||||
- `yolov8[ns]-oiv7_320x320_edgetpu.tflite` -- model files that have been trained using the google open images v7 dataset (601 classes)
|
||||
- `labels.txt` and `labels-frigate.txt` -- full and aggregated labels for the coco dataset models
|
||||
- `labels-oiv7.txt` and `labels-oiv7-frigate.txt` -- labels for the oiv7 dataset models
|
||||
|
||||
The aggregated label files contain renamed labels leaving only `person`, `vehicle`, `animal` and `bird` classes. The oiv7 trained models contain 601 classes and so are difficult to configure manually -- using aggregate labels is recommended.
|
||||
|
||||
Larger models (of `m` and `l` size and also at `640x640` resolution) can be found at https://github.com/harakas/models/releases/tag/yolov8.1-1.1/ but have to be installed manually.
|
||||
|
||||
The oiv7 models have been trained using a larger google open images v7 dataset. They also contain a lot more detection classes (over 600) so using aggregate label files is recommended. The large number of classes leads to lower baseline for detection probability values and also for higher resource consumption (they are slower to evaluate).
|
||||
|
||||
#### Configuration
|
||||
|
||||
```yaml
|
||||
model:
|
||||
labelmap_path: /config/model_cache/yolov8/labels.txt
|
||||
model_type: yolov8
|
||||
detectors:
|
||||
coral:
|
||||
type: edgetpu
|
||||
device: usb
|
||||
model:
|
||||
path: /config/model_cache/yolov8/yolov8n_320x320_edgetpu.tflite
|
||||
```
|
||||
|
||||
## OpenVINO Detector
|
||||
|
||||
The OpenVINO detector type runs an OpenVINO IR model on AMD and Intel CPUs, Intel GPUs and Intel VPU hardware. To configure an OpenVINO detector, set the `"type"` attribute to `"openvino"`.
|
||||
|
||||
@ -6,7 +6,6 @@ from typing_extensions import Literal
|
||||
|
||||
from frigate.detectors.detection_api import DetectionApi
|
||||
from frigate.detectors.detector_config import BaseDetectorConfig
|
||||
from frigate.detectors.util import yolov8_postprocess
|
||||
|
||||
try:
|
||||
from tflite_runtime.interpreter import Interpreter, load_delegate
|
||||
@ -58,26 +57,9 @@ class EdgeTpuTfl(DetectionApi):
|
||||
self.model_type = detector_config.model.model_type
|
||||
|
||||
def detect_raw(self, tensor_input):
|
||||
if self.model_type == "yolov8":
|
||||
scale, zero_point = self.tensor_input_details[0]["quantization"]
|
||||
tensor_input = (
|
||||
(tensor_input - scale * zero_point * 255) * (1.0 / (scale * 255))
|
||||
).astype(self.tensor_input_details[0]["dtype"])
|
||||
|
||||
self.interpreter.set_tensor(self.tensor_input_details[0]["index"], tensor_input)
|
||||
self.interpreter.invoke()
|
||||
|
||||
if self.model_type == "yolov8":
|
||||
scale, zero_point = self.tensor_output_details[0]["quantization"]
|
||||
tensor_output = self.interpreter.get_tensor(
|
||||
self.tensor_output_details[0]["index"]
|
||||
)
|
||||
tensor_output = (tensor_output.astype(np.float32) - zero_point) * scale
|
||||
model_input_shape = self.tensor_input_details[0]["shape"]
|
||||
tensor_output[:, [0, 2]] *= model_input_shape[2]
|
||||
tensor_output[:, [1, 3]] *= model_input_shape[1]
|
||||
return yolov8_postprocess(model_input_shape, tensor_output)
|
||||
|
||||
boxes = self.interpreter.tensor(self.tensor_output_details[0]["index"])()[0]
|
||||
class_ids = self.interpreter.tensor(self.tensor_output_details[1]["index"])()[0]
|
||||
scores = self.interpreter.tensor(self.tensor_output_details[2]["index"])()[0]
|
||||
|
||||
@ -6,7 +6,7 @@ from typing_extensions import Literal
|
||||
|
||||
from frigate.detectors.detection_api import DetectionApi
|
||||
from frigate.detectors.detector_config import BaseDetectorConfig
|
||||
from frigate.detectors.util import preprocess, yolov8_postprocess
|
||||
from frigate.detectors.util import preprocess
|
||||
|
||||
logger = logging.getLogger(__name__)
|
||||
|
||||
@ -31,23 +31,6 @@ class ONNXDetector(DetectionApi):
|
||||
)
|
||||
raise
|
||||
|
||||
assert (
|
||||
detector_config.model.model_type == "yolov8"
|
||||
), "ONNX: detector_config.model.model_type: only yolov8 supported"
|
||||
assert (
|
||||
detector_config.model.input_tensor == "nhwc"
|
||||
), "ONNX: detector_config.model.input_tensor: only nhwc supported"
|
||||
if detector_config.model.input_pixel_format != "rgb":
|
||||
logger.warn(
|
||||
"ONNX: detector_config.model.input_pixel_format: should be 'rgb' for yolov8, but '{detector_config.model.input_pixel_format}' specified!"
|
||||
)
|
||||
|
||||
assert detector_config.model.path is not None, (
|
||||
"ONNX: No model.path configured, please configure model.path and model.labelmap_path; some suggestions: "
|
||||
+ ", ".join(glob.glob("/config/model_cache/yolov8/*.onnx"))
|
||||
+ " and "
|
||||
+ ", ".join(glob.glob("/config/model_cache/yolov8/*_labels.txt"))
|
||||
)
|
||||
|
||||
path = detector_config.model.path
|
||||
logger.info(f"ONNX: loading {detector_config.model.path}")
|
||||
@ -62,4 +45,4 @@ class ONNXDetector(DetectionApi):
|
||||
|
||||
tensor_output = self.model.run(None, {model_input_name: tensor_input})[0]
|
||||
|
||||
return yolov8_postprocess(model_input_shape, tensor_output)
|
||||
raise Exception("No models are currently supported via onnx. See the docs for more info.")
|
||||
|
||||
@ -1,5 +1,4 @@
|
||||
import ctypes
|
||||
import glob
|
||||
import logging
|
||||
import os
|
||||
import subprocess
|
||||
@ -11,7 +10,7 @@ from typing_extensions import Literal
|
||||
|
||||
from frigate.detectors.detection_api import DetectionApi
|
||||
from frigate.detectors.detector_config import BaseDetectorConfig
|
||||
from frigate.detectors.util import preprocess, yolov8_postprocess
|
||||
from frigate.detectors.util import preprocess
|
||||
|
||||
logger = logging.getLogger(__name__)
|
||||
|
||||
@ -75,27 +74,6 @@ class ROCmDetector(DetectionApi):
|
||||
logger.error("AMD/ROCm: module loading failed, missing ROCm environment?")
|
||||
raise
|
||||
|
||||
if detector_config.conserve_cpu:
|
||||
logger.info("AMD/ROCm: switching HIP to blocking mode to conserve CPU")
|
||||
ctypes.CDLL("/opt/rocm/lib/libamdhip64.so").hipSetDeviceFlags(4)
|
||||
assert (
|
||||
detector_config.model.model_type == "yolov8"
|
||||
), "AMD/ROCm: detector_config.model.model_type: only yolov8 supported"
|
||||
assert (
|
||||
detector_config.model.input_tensor == "nhwc"
|
||||
), "AMD/ROCm: detector_config.model.input_tensor: only nhwc supported"
|
||||
if detector_config.model.input_pixel_format != "rgb":
|
||||
logger.warn(
|
||||
"AMD/ROCm: detector_config.model.input_pixel_format: should be 'rgb' for yolov8, but '{detector_config.model.input_pixel_format}' specified!"
|
||||
)
|
||||
|
||||
assert detector_config.model.path is not None, (
|
||||
"No model.path configured, please configure model.path and model.labelmap_path; some suggestions: "
|
||||
+ ", ".join(glob.glob("/config/model_cache/yolov8/*.onnx"))
|
||||
+ " and "
|
||||
+ ", ".join(glob.glob("/config/model_cache/yolov8/*_labels.txt"))
|
||||
)
|
||||
|
||||
path = detector_config.model.path
|
||||
mxr_path = os.path.splitext(path)[0] + ".mxr"
|
||||
if path.endswith(".mxr"):
|
||||
@ -140,4 +118,4 @@ class ROCmDetector(DetectionApi):
|
||||
addr, shape=detector_result.get_shape().lens()
|
||||
)
|
||||
|
||||
return yolov8_postprocess(model_input_shape, tensor_output)
|
||||
raise Exception("No models are currently supported for rocm. See the docs for more info.")
|
||||
|
||||
@ -34,50 +34,3 @@ def preprocess(tensor_input, model_input_shape, model_input_element_type):
|
||||
None,
|
||||
swapRB=False,
|
||||
)
|
||||
|
||||
|
||||
def yolov8_postprocess(
|
||||
model_input_shape,
|
||||
tensor_output,
|
||||
box_count=20,
|
||||
score_threshold=0.5,
|
||||
nms_threshold=0.5,
|
||||
):
|
||||
model_box_count = tensor_output.shape[2]
|
||||
probs = tensor_output[0, 4:, :]
|
||||
all_ids = np.argmax(probs, axis=0)
|
||||
all_confidences = probs.T[np.arange(model_box_count), all_ids]
|
||||
all_boxes = tensor_output[0, 0:4, :].T
|
||||
mask = all_confidences > score_threshold
|
||||
class_ids = all_ids[mask]
|
||||
confidences = all_confidences[mask]
|
||||
cx, cy, w, h = all_boxes[mask].T
|
||||
|
||||
if model_input_shape[3] == 3:
|
||||
scale_y, scale_x = 1 / model_input_shape[1], 1 / model_input_shape[2]
|
||||
else:
|
||||
scale_y, scale_x = 1 / model_input_shape[2], 1 / model_input_shape[3]
|
||||
detections = np.stack(
|
||||
(
|
||||
class_ids,
|
||||
confidences,
|
||||
scale_y * (cy - h / 2),
|
||||
scale_x * (cx - w / 2),
|
||||
scale_y * (cy + h / 2),
|
||||
scale_x * (cx + w / 2),
|
||||
),
|
||||
axis=1,
|
||||
)
|
||||
if detections.shape[0] > box_count:
|
||||
# if too many detections, do nms filtering to suppress overlapping boxes
|
||||
boxes = np.stack((cx - w / 2, cy - h / 2, w, h), axis=1)
|
||||
indexes = cv2.dnn.NMSBoxes(boxes, confidences, score_threshold, nms_threshold)
|
||||
detections = detections[indexes]
|
||||
# if still too many, trim the rest by confidence
|
||||
if detections.shape[0] > box_count:
|
||||
detections = detections[
|
||||
np.argpartition(detections[:, 1], -box_count)[-box_count:]
|
||||
]
|
||||
detections = detections.copy()
|
||||
detections.resize((box_count, 6))
|
||||
return detections
|
||||
|
||||
Loading…
Reference in New Issue
Block a user