Comments (2)
Everything is normal. You have finished fine-tuning the Segmenter, and you can now use this weight for subsequent online and offline DVIS training. Currently, CTVIS and MinVIS only support VIS inference. They are only used to provide pre-trained Segmenter for DVIS. Evaluation of VSS and VPS can only be conducted when running DVIS online or offline.
from dvis_plus.
Thanks for the response!
In the meanwhile I found a rough way to adapt the vps_eval.py process function to allow evaluating segmentation models directly without needing to train DVIS on top of it. This is useful if one wants to judge the quality of this segmentation model before starting the additional training on top. I am posting it here for future reference.
def process(self, inputs, outputs):
"""
save panoptic segmentation result as an image
"""
assert len(inputs) == 1, "More than one inputs are loaded for inference!"
color_generator = IdGenerator(self._metadata.categories)
# visualization of only panoptic segmentation results like this
if 'segments_infos' not in outputs:
mask_pred = torch.stack(outputs['pred_masks']).float()
labels = torch.asarray(outputs['pred_labels'])
scores = torch.asarray(outputs['pred_scores'])
# filter out the background prediction
keep = labels.ne(124) & (scores > 0.8)
cur_scores = scores[keep]
cur_classes = labels[keep]
cur_masks = mask_pred[keep]
# interpolation to original image size
cur_masks = F.interpolate(
cur_masks, size=(736, 1280), mode="bilinear", align_corners=False
)
cur_masks = cur_masks[:, :, :720, :1280].sigmoid()
cur_masks = F.interpolate(
cur_masks, size=(720, 1280), mode="bilinear", align_corners=False
)
cur_prob_masks = cur_scores.view(-1, 1, 1, 1).to(cur_masks.device) * cur_masks
# initial panoptic_seg and segments infos
h, w = cur_masks.shape[-2:]
panoptic_seg = torch.zeros((cur_masks.size(1), h, w), dtype=torch.int32, device=cur_masks.device)
segments_infos = []
out_ids = []
current_segment_id = 0
if cur_masks.shape[0] == 0:
# We didn't detect any mask
outputs = {
"image_size": (720, 1280),
"pred_masks": panoptic_seg.cpu(),
"segments_infos": segments_infos,
"pred_ids": out_ids,
"task": "vps",
}
else:
# take argmax
cur_mask_ids = cur_prob_masks.argmax(0) # (t, h, w)
stuff_memory_list = {}
for k in range(cur_classes.shape[0]):
pred_class = cur_classes[k].item()
isthing = pred_class < 58
# filter out the unstable segmentation results
mask_area = (cur_mask_ids == k).sum().item()
original_area = (cur_masks[k] >= 0.5).sum().item()
mask = (cur_mask_ids == k) & (cur_masks[k] >= 0.5)
if mask_area > 0 and original_area > 0 and mask.sum().item() > 0:
if mask_area / original_area < 0.01: # 0.8:
continue
# merge stuff regions
if not isthing:
if int(pred_class) in stuff_memory_list.keys():
panoptic_seg[mask] = stuff_memory_list[int(pred_class)]
continue
else:
stuff_memory_list[int(pred_class)] = current_segment_id + 1
current_segment_id += 1
panoptic_seg[mask] = current_segment_id
segments_infos.append(
{
"id": current_segment_id,
"isthing": bool(isthing),
"category_id": int(pred_class),
}
)
outputs = {
"image_size": (720, 1280),
"pred_masks": panoptic_seg.cpu(),
"segments_infos": segments_infos,
"pred_ids": out_ids,
"task": "vps",
}
video_id = inputs[0]["video_id"]
image_names = [inputs[0]['file_names'][idx] for idx in inputs[0]["frame_idx"]]
img_shape = outputs['image_size']
pan_seg_result = outputs['pred_masks']
segments_infos = outputs['segments_infos']
segments_infos_ = []
...Rest of the code....
from dvis_plus.
Related Issues (14)
- Download model weights baidu HOT 3
- Questions about results.json
- Cannot load pretrained weight HOT 2
- Temporal Refiner in Online Mode HOT 2
- VIPseg panoptic_gt_VIPSeg_test.json
- demo.py: error: unrecognized arguments: --merge: command not found HOT 2
- eval: KeyError: 'video_id` / demo: AttributeError: 'list' object has no attribute 'to' HOT 4
- about the number of gpus or IMS_PER_BATCH HOT 9
- infer
- infer HOT 1
- error running HOT 1
- Error while Evaluation
- Custom dataset
Recommend Projects
-
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
-
Vue.js
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
-
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
-
TensorFlow
An Open Source Machine Learning Framework for Everyone
-
Django
The Web framework for perfectionists with deadlines.
-
Laravel
A PHP framework for web artisans
-
D3
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
-
Recommend Topics
-
javascript
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
-
web
Some thing interesting about web. New door for the world.
-
server
A server is a program made to process requests and deliver data to clients.
-
Machine learning
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
-
Visualization
Some thing interesting about visualization, use data art
-
Game
Some thing interesting about game, make everyone happy.
Recommend Org
-
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
-
Microsoft
Open source projects and samples from Microsoft.
-
Google
Google ❤️ Open Source for everyone.
-
Alibaba
Alibaba Open Source for everyone
-
D3
Data-Driven Documents codes.
-
Tencent
China tencent open source team.
from dvis_plus.