refactor(retrieval): polish retrieval.py

ppcls/engine/evaluation/retrieval.py

@@ -16,137 +16,132 @@ from __future__ import division
 from __future__ import print_function
 
 import platform
-from typing import Optional
 
 import numpy as np
 import paddle
 from ppcls.engine.train.utils import type_name
 from ppcls.utils import logger
+from ppcls.utils import all_gather
 
 
 def retrieval_eval(engine, epoch_id=0):
     engine.model.eval()
-    # step1. build query & gallery
+    # step1. prepare query and gallery features
     if engine.gallery_query_dataloader is not None:
-        gallery_feas, gallery_img_id, gallery_unique_id = cal_feature(
-            engine, name='gallery_query')
-        query_feas, query_img_id, query_unique_id = gallery_feas, gallery_img_id, gallery_unique_id
+        gallery_feas, gallery_label_id, gallery_camera_id = compute_feature(
+            engine, "gallery_query")
+        query_feas, query_label_id, query_camera_id = gallery_feas, gallery_label_id, gallery_camera_id
     else:
-        gallery_feas, gallery_img_id, gallery_unique_id = cal_feature(
-            engine, name='gallery')
-        query_feas, query_img_id, query_unique_id = cal_feature(
-            engine, name='query')
-
-    # step2. split data into blocks so as to save memory
-    sim_block_size = engine.config["Global"].get("sim_block_size", 64)
-    sections = [sim_block_size] * (len(query_feas) // sim_block_size)
-    if len(query_feas) % sim_block_size:
-        sections.append(len(query_feas) % sim_block_size)
-
-    fea_blocks = paddle.split(query_feas, num_or_sections=sections)
-    if query_unique_id is not None:
-        query_unique_id_blocks = paddle.split(
-            query_unique_id, num_or_sections=sections)
-    query_img_id_blocks = paddle.split(query_img_id, num_or_sections=sections)
+        gallery_feas, gallery_label_id, gallery_camera_id = compute_feature(
+            engine, "gallery")
+        query_feas, query_label_id, query_camera_id = compute_feature(engine,
+                                                                      "query")
+
+    # step2. split features into feature blocks for saving memory
+    block_size = engine.config["Global"].get("sim_block_size", 64)
+    sections = [block_size] * (len(query_feas) // block_size)
+    if len(query_feas) % block_size > 0:
+        sections.append(len(query_feas) % block_size)
+
+    query_feas_blocks = paddle.split(query_feas, sections)
+    query_camera_id_blocks = (paddle.split(query_camera_id, sections)
+                              if query_camera_id is not None else None)
+    query_label_id_blocks = paddle.split(query_label_id, sections)
     metric_key = None
 
-    # step3. do evaluation
+    # step3. compute metric
     if engine.eval_loss_func is None:
         metric_dict = {metric_key: 0.}
     else:
-        # do evaluation with re-ranking(k-reciprocal)
-        reranking_flag = engine.config['Global'].get('re_ranking', False)
-        logger.info(f"re_ranking={reranking_flag}")
-        metric_dict = dict()
-        if reranking_flag:
-            # set the order from small to large
-            for i in range(len(engine.eval_metric_func.metric_func_list)):
-                if hasattr(engine.eval_metric_func.metric_func_list[i], 'descending') \
-                        and engine.eval_metric_func.metric_func_list[i].descending is True:
-                    engine.eval_metric_func.metric_func_list[
-                        i].descending = False
+        use_reranking = engine.config["Global"].get("re_ranking", False)
+        logger.info(f"re_ranking={use_reranking}")
+        metric_dict = {}
+        if use_reranking:
+            for _, metric_func in enumerate(
+                    engine.eval_metric_func.metric_func_list):
+                if hasattr(metric_func,
+                           "descending") and metric_func.descending is True:
+                    metric_func.descending = False
                     logger.warning(
-                        f"re_ranking=True,{type_name(engine.eval_metric_func.metric_func_list[i])}.descending has been set to False"
+                        f"re_ranking=True, set {type_name(metric_func)}.descending set to False"
                     )
-
-            # compute distance matrix(The smaller the value, the more similar)
-            distmat = re_ranking(
-                query_feas, gallery_feas, k1=20, k2=6, lambda_value=0.3)
-
-            # compute keep mask
-            unique_id_mask = (query_unique_id != gallery_unique_id.t())
-            image_id_mask = (query_img_id != gallery_img_id.t())
-            keep_mask = paddle.logical_or(image_id_mask, unique_id_mask)
-
-            # set inf(1e9) distance to those exist in gallery
-            distmat = distmat * keep_mask.astype("float32")
-            inf_mat = (paddle.logical_not(keep_mask).astype("float32")) * 1e20
+            # compute distance matrix
+            distmat = compute_re_ranking_dist(
+                query_feas, gallery_feas, engine.config["Global"].get(
+                    "feature_normalize", True), 20, 6, 0.3)
+
+            # exclude illegal distance
+            camera_id_mask = query_camera_id != gallery_camera_id.t()
+            image_id_mask = query_label_id != gallery_label_id.t()
+            keep_mask = paddle.logical_or(image_id_mask, camera_id_mask)
+            distmat = distmat * keep_mask.astype(query_feas.dtype)
+            inf_mat = (
+                paddle.logical_not(keep_mask).astype(query_feas.dtype)) * (
+                    distmat.max() + 1)
             distmat = distmat + inf_mat
 
-            # compute metric
-            metric_tmp = engine.eval_metric_func(distmat, query_img_id,
-                                                 gallery_img_id, keep_mask)
-            for key in metric_tmp:
-                metric_dict[key] = metric_tmp[key]
+            metric_block = engine.eval_metric_func(distmat, query_label_id,
+                                                   gallery_label_id, keep_mask)
+            for key in metric_block:
+                metric_dict[key] = metric_block[key]
         else:
-            # do evaluation without re-ranking
-            for block_idx, block_fea in enumerate(fea_blocks):
-                similarity_matrix = paddle.matmul(
-                    block_fea, gallery_feas, transpose_y=True)  # [n,m]
-                if query_unique_id is not None:
-                    query_unique_id_block = query_unique_id_blocks[block_idx]
-                    unique_id_mask = (
-                        query_unique_id_block != gallery_unique_id.t())
-
-                    query_img_id_block = query_img_id_blocks[block_idx]
-                    image_id_mask = (query_img_id_block != gallery_img_id.t())
+            for block_idx, block_fea in enumerate(query_feas_blocks):
+                distmat = paddle.matmul(
+                    block_fea, gallery_feas, transpose_y=True)
+                if query_camera_id is not None:
+                    query_camera_id_block = query_camera_id_blocks[block_idx]
+                    camera_id_mask = query_camera_id_block != gallery_camera_id.t(
+                    )
+
+                    query_label_id_block = query_label_id_blocks[block_idx]
+                    image_id_mask = query_label_id_block != gallery_label_id.t(
+                    )
 
                     keep_mask = paddle.logical_or(image_id_mask,
-                                                  unique_id_mask)
-                    similarity_matrix = similarity_matrix * keep_mask.astype(
-                        "float32")
+                                                  camera_id_mask)
+                    distmat = distmat * keep_mask.astype("float32")
                 else:
                     keep_mask = None
 
-                metric_tmp = engine.eval_metric_func(
-                    similarity_matrix, query_img_id_blocks[block_idx],
-                    gallery_img_id, keep_mask)
+                metric_block = engine.eval_metric_func(
+                    distmat, query_label_id_blocks[block_idx],
+                    gallery_label_id, keep_mask)
 
-                for key in metric_tmp:
+                for key in metric_block:
                     if key not in metric_dict:
-                        metric_dict[key] = metric_tmp[key] * block_fea.shape[
+                        metric_dict[key] = metric_block[key] * block_fea.shape[
                             0] / len(query_feas)
                     else:
-                        metric_dict[key] += metric_tmp[key] * block_fea.shape[
-                            0] / len(query_feas)
+                        metric_dict[key] += metric_block[
+                            key] * block_fea.shape[0] / len(query_feas)
 
     metric_info_list = []
     for key in metric_dict:
+        metric_info_list.append(f"{key}: {metric_dict[key]:.5f}")
         if metric_key is None:
             metric_key = key
-        metric_info_list.append("{}: {:.5f}".format(key, metric_dict[key]))
     metric_msg = ", ".join(metric_info_list)
-    logger.info("[Eval][Epoch {}][Avg]{}".format(epoch_id, metric_msg))
+    logger.info(f"[Eval][Epoch {epoch_id}][Avg]{metric_msg}")
 
     return metric_dict[metric_key]
 
 
-def cal_feature(engine, name='gallery'):
-    has_unique_id = False
-    all_unique_id = None
+def compute_feature(engine, name="gallery"):
+    has_camera_id = False
+    all_camera_id = None
 
-    if name == 'gallery':
+    if name == "gallery":
         dataloader = engine.gallery_dataloader
-    elif name == 'query':
+    elif name == "query":
         dataloader = engine.query_dataloader
-    elif name == 'gallery_query':
+    elif name == "gallery_query":
         dataloader = engine.gallery_query_dataloader
     else:
         raise RuntimeError("Only support gallery or query dataset")
 
     batch_feas_list = []
-    img_id_list = []
-    unique_id_list = []
+    label_id_list = []
+    camera_id_list = []
     max_iter = len(dataloader) - 1 if platform.system() == "Windows" else len(
         dataloader)
     for idx, batch in enumerate(dataloader):  # load is very time-consuming
@@ -160,7 +155,7 @@ def cal_feature(engine, name='gallery'):
         batch = [paddle.to_tensor(x) for x in batch]
         batch[1] = batch[1].reshape([-1, 1]).astype("int64")
         if len(batch) == 3:
-            has_unique_id = True
+            has_camera_id = True
             batch[2] = batch[2].reshape([-1, 1]).astype("int64")
         if engine.amp and engine.amp_eval:
             with paddle.amp.auto_cast(
@@ -183,158 +178,160 @@ def cal_feature(engine, name='gallery'):
             # use backbone's output as features
             batch_feas = out["backbone"]
 
-        # do norm
+        # do norm(optinal)
         if engine.config["Global"].get("feature_normalize", True):
-            feas_norm = paddle.sqrt(
+            batch_feas_norm = paddle.sqrt(
                 paddle.sum(paddle.square(batch_feas), axis=1, keepdim=True))
-            batch_feas = paddle.divide(batch_feas, feas_norm)
+            batch_feas = paddle.divide(batch_feas, batch_feas_norm)
 
-        # do binarize
+        # do binarize(optinal)
         if engine.config["Global"].get("feature_binarize") == "round":
             batch_feas = paddle.round(batch_feas).astype("float32") * 2.0 - 1.0
-
-        if engine.config["Global"].get("feature_binarize") == "sign":
+        elif engine.config["Global"].get("feature_binarize") == "sign":
             batch_feas = paddle.sign(batch_feas).astype("float32")
 
         if paddle.distributed.get_world_size() > 1:
-            batch_feas_gather = []
-            img_id_gather = []
-            unique_id_gather = []
-            paddle.distributed.all_gather(batch_feas_gather, batch_feas)
-            paddle.distributed.all_gather(img_id_gather, batch[1])
-            batch_feas_list.append(paddle.concat(batch_feas_gather))
-            img_id_list.append(paddle.concat(img_id_gather))
-            if has_unique_id:
-                paddle.distributed.all_gather(unique_id_gather, batch[2])
-                unique_id_list.append(paddle.concat(unique_id_gather))
+            batch_feas_list.append(all_gather(batch_feas))
+            label_id_list.append(all_gather(batch[1]))
+            if has_camera_id:
+                camera_id_list.append(all_gather(batch[2]))
         else:
             batch_feas_list.append(batch_feas)
-            img_id_list.append(batch[1])
-            if has_unique_id:
-                unique_id_list.append(batch[2])
+            label_id_list.append(batch[1])
+            if has_camera_id:
+                camera_id_list.append(batch[2])
 
     if engine.use_dali:
         dataloader.reset()
 
     all_feas = paddle.concat(batch_feas_list)
-    all_img_id = paddle.concat(img_id_list)
-    if has_unique_id:
-        all_unique_id = paddle.concat(unique_id_list)
+    all_label_id = paddle.concat(label_id_list)
+    if has_camera_id:
+        all_camera_id = paddle.concat(camera_id_list)
 
-    # just for DistributedBatchSampler issue: repeat sampling
+    # discard redundant padding sample(s) at the end
     total_samples = len(
         dataloader.dataset) if not engine.use_dali else dataloader.size
     all_feas = all_feas[:total_samples]
-    all_img_id = all_img_id[:total_samples]
-    if has_unique_id:
-        all_unique_id = all_unique_id[:total_samples]
+    all_label_id = all_label_id[:total_samples]
+    if has_camera_id:
+        all_camera_id = all_camera_id[:total_samples]
 
-    logger.info("Build {} done, all feat shape: {}, begin to eval..".format(
-        name, all_feas.shape))
-    return all_feas, all_img_id, all_unique_id
+    logger.info(f"Build {name} done, all feat shape: {all_feas.shape}")
+    return all_feas, all_label_id, all_camera_id
 
 
-def re_ranking(query_feas: paddle.Tensor,
-               gallery_feas: paddle.Tensor,
-               k1: int=20,
-               k2: int=6,
-               lambda_value: int=0.5,
-               local_distmat: Optional[np.ndarray]=None,
-               only_local: bool=False) -> paddle.Tensor:
-    """re-ranking, most computed with numpy
-
-    code heavily based on
-    https://github.com/michuanhaohao/reid-strong-baseline/blob/3da7e6f03164a92e696cb6da059b1cd771b0346d/utils/reid_metric.py
+def k_reciprocal_neighbor(rank: np.ndarray, p: int, k: int) -> np.ndarray:
+    """Implementation of k-reciprocal nearest neighbors, i.e. R(p, k)
 
     Args:
-        query_feas (paddle.Tensor): query features, [num_query, num_features]
-        gallery_feas (paddle.Tensor): gallery features, [num_gallery, num_features]
-        k1 (int, optional): k1. Defaults to 20.
-        k2 (int, optional): k2. Defaults to 6.
-        lambda_value (int, optional): lambda. Defaults to 0.5.
-        local_distmat (Optional[np.ndarray], optional): local_distmat. Defaults to None.
-        only_local (bool, optional): only_local. Defaults to False.
+        rank (np.ndarray): Rank mat with shape of [N, N].
+        p (int): Probe index.
+        k (int): Parameter k for k-reciprocal nearest neighbors algorithm.
 
     Returns:
-        paddle.Tensor: final_dist matrix after re-ranking, [num_query, num_gallery]
+        np.ndarray: K-reciprocal nearest neighbors of probe p with shape of [M, ].
     """
-    query_num = query_feas.shape[0]
-    all_num = query_num + gallery_feas.shape[0]
-    if only_local:
-        original_dist = local_distmat
-    else:
-        feat = paddle.concat([query_feas, gallery_feas])
-        logger.info('using GPU to compute original distance')
+    # use k+1 for excluding probe index itself
+    forward_k_neigh_index = rank[p, :k + 1]
+    backward_k_neigh_index = rank[forward_k_neigh_index, :k + 1]
+    candidate = np.where(backward_k_neigh_index == p)[0]
+    return forward_k_neigh_index[candidate]
+
+
+def compute_re_ranking_dist(query_feas: paddle.Tensor,
+                            gallery_feas: paddle.Tensor,
+                            feature_normed: bool=True,
+                            k1: int=20,
+                            k2: int=6,
+                            lamb: float=0.5) -> paddle.Tensor:
+    """
+    Re-ranking Person Re-identification with k-reciprocal Encoding
+    Reference: https://arxiv.org/abs/1701.08398
+    Code refernence: https://github.com/michuanhaohao/reid-strong-baseline/blob/master/utils/re_ranking.py
 
-        # L2 distance
-        distmat = paddle.pow(feat, 2).sum(axis=1, keepdim=True).expand([all_num, all_num]) + \
-            paddle.pow(feat, 2).sum(axis=1, keepdim=True).expand([all_num, all_num]).t()
-        distmat = distmat.addmm(x=feat, y=feat.t(), alpha=-2.0, beta=1.0)
+    Args:
+        query_feas (paddle.Tensor): Query features with shape of [num_query, feature_dim].
+        gallery_feas (paddle.Tensor):  Gallery features with shape of [num_gallery, feature_dim].
+        feature_normed (bool, optional):  Whether input features are normalized.
+        k1 (int, optional): Parameter for K-reciprocal nearest neighbors. Defaults to 20.
+        k2 (int, optional): Parameter for K-nearest neighbors. Defaults to 6.
+        lamb (float, optional): Penalty factor. Defaults to 0.5.
 
-        original_dist = distmat.cpu().numpy()
-        del feat
-        if local_distmat is not None:
-            original_dist = original_dist + local_distmat
+    Returns:
+        paddle.Tensor: (1 - lamb) x Dj + lamb x D, with shape of [num_query, num_gallery].
+    """
+    num_query = query_feas.shape[0]
+    num_gallery = gallery_feas.shape[0]
+    num_all = num_query + num_gallery
+    feat = paddle.concat([query_feas, gallery_feas], 0)
+    logger.info("Using GPU to compute original distance matrix")
+
+    # use L2 distance
+    if feature_normed:
+        original_dist = 2 - 2 * paddle.matmul(feat, feat, transpose_y=True)
+    else:
+        original_dist = paddle.pow(feat, 2).sum(axis=1, keepdim=True).expand([num_all, num_all]) + \
+            paddle.pow(feat, 2).sum(axis=1, keepdim=True).expand([num_all, num_all]).t()
+        original_dist = original_dist.addmm(
+            x=feat, y=feat.t(), alpha=-2.0, beta=1.0)
+    original_dist = original_dist.numpy()
+    del feat
 
-    gallery_num = original_dist.shape[0]
     original_dist = np.transpose(original_dist / np.max(original_dist, axis=0))
     V = np.zeros_like(original_dist).astype(np.float16)
-    initial_rank = np.argsort(original_dist).astype(np.int32)
-    logger.info('starting re_ranking')
-    for i in range(all_num):
-        # k-reciprocal neighbors
-        forward_k_neigh_index = initial_rank[i, :k1 + 1]
-        backward_k_neigh_index = initial_rank[forward_k_neigh_index, :k1 + 1]
-        fi = np.where(backward_k_neigh_index == i)[0]
-        k_reciprocal_index = forward_k_neigh_index[fi]
-        k_reciprocal_expansion_index = k_reciprocal_index
-        for j in range(len(k_reciprocal_index)):
-            candidate = k_reciprocal_index[j]
-            candidate_forward_k_neigh_index = initial_rank[candidate, :int(
-                np.around(k1 / 2)) + 1]
-            candidate_backward_k_neigh_index = initial_rank[
-                candidate_forward_k_neigh_index, :int(np.around(k1 / 2)) + 1]
-            fi_candidate = np.where(
-                candidate_backward_k_neigh_index == candidate)[0]
-            candidate_k_reciprocal_index = candidate_forward_k_neigh_index[
-                fi_candidate]
-            if len(
-                    np.intersect1d(candidate_k_reciprocal_index,
-                                   k_reciprocal_index)) > 2 / 3 * len(
-                                       candidate_k_reciprocal_index):
-                k_reciprocal_expansion_index = np.append(
-                    k_reciprocal_expansion_index, candidate_k_reciprocal_index)
-
-        k_reciprocal_expansion_index = np.unique(k_reciprocal_expansion_index)
-        weight = np.exp(-original_dist[i, k_reciprocal_expansion_index])
-        V[i, k_reciprocal_expansion_index] = weight / np.sum(weight)
-    original_dist = original_dist[:query_num, ]
-    if k2 != 1:
+    initial_rank = np.argpartition(original_dist, range(1, k1 + 1))
+    logger.info("Start re-ranking...")
+
+    for p in range(num_all):
+        # compute R(p,k1)
+        p_k_reciprocal_ind = k_reciprocal_neighbor(initial_rank, p, k1)
+
+        # compute R*(p,k1)=R(p,k1)∪R(q,k1/2)
+        # s.t. |R(p,k1)∩R(q,k1/2)|>=2/3|R(q,k1/2)|, ∀q∈R(p,k1)
+        p_k_reciprocal_exp_ind = p_k_reciprocal_ind
+        for _, q in enumerate(p_k_reciprocal_ind):
+            q_k_reciprocal_ind = k_reciprocal_neighbor(initial_rank, q,
+                                                       int(np.around(k1 / 2)))
+            if len(np.intersect1d(p_k_reciprocal_ind, q_k_reciprocal_ind)
+                   ) > 2 / 3 * len(q_k_reciprocal_ind):
+                p_k_reciprocal_exp_ind = np.append(p_k_reciprocal_exp_ind,
+                                                   q_k_reciprocal_ind)
+        p_k_reciprocal_exp_ind = np.unique(p_k_reciprocal_exp_ind)
+
+        # reweight distance using gaussian kernel
+        weight = np.exp(-original_dist[p, p_k_reciprocal_exp_ind])
+        V[p, p_k_reciprocal_exp_ind] = weight / np.sum(weight)
+
+    # local query expansion
+    original_dist = original_dist[:num_query, ]
+    if k2 > 1:
         V_qe = np.zeros_like(V, dtype=np.float16)
-        for i in range(all_num):
-            V_qe[i, :] = np.mean(V[initial_rank[i, :k2], :], axis=0)
+        for p in range(num_all):
+            V_qe[p, :] = np.mean(V[initial_rank[p, :k2], :], axis=0)
         V = V_qe
         del V_qe
     del initial_rank
+
+    # cache k-reciprocal sets which contains gj
     invIndex = []
-    for i in range(gallery_num):
-        invIndex.append(np.where(V[:, i] != 0)[0])
+    for gj in range(num_all):
+        invIndex.append(np.nonzero(V[:, gj])[0])
 
     jaccard_dist = np.zeros_like(original_dist, dtype=np.float16)
-    for i in range(query_num):
-        temp_min = np.zeros(shape=[1, gallery_num], dtype=np.float16)
-        indNonZero = np.where(V[i, :] != 0)[0]
-        indImages = [invIndex[ind] for ind in indNonZero]
-        for j in range(len(indNonZero)):
-            temp_min[0, indImages[j]] = temp_min[0, indImages[j]] + np.minimum(
-                V[i, indNonZero[j]], V[indImages[j], indNonZero[j]])
-        jaccard_dist[i] = 1 - temp_min / (2 - temp_min)
-
-    final_dist = jaccard_dist * (1 - lambda_value
-                                 ) + original_dist * lambda_value
+    for p in range(num_query):
+        sum_min = np.zeros(shape=[1, num_all], dtype=np.float16)
+        gj_ind = np.nonzero(V[p, :])[0]
+        gj_ind_inv = [invIndex[gj] for gj in gj_ind]
+        for j, gj in enumerate(gj_ind):
+            gi = gj_ind_inv[j]
+            sum_min[0, gi] += np.minimum(V[p, gj], V[gi, gj])
+        jaccard_dist[p] = 1 - sum_min / (2 - sum_min)
+
+    final_dist = jaccard_dist * (1 - lamb) + original_dist * lamb
     del original_dist
     del V
     del jaccard_dist
-    final_dist = final_dist[:query_num, query_num:]
+    final_dist = final_dist[:num_query, num_query:]
     final_dist = paddle.to_tensor(final_dist)
     return final_dist

ppcls/utils/__init__.py

@@ -13,15 +13,16 @@
 # limitations under the License.
 
 from . import logger
+from . import metrics
 from . import misc
 from . import model_zoo
-from . import metrics
 
-from .save_load import init_model, save_model
 from .config import get_config
-from .misc import AverageMeter
-from .metrics import multi_hot_encode
-from .metrics import hamming_distance
+from .dist_utils import all_gather
 from .metrics import accuracy_score
-from .metrics import precision_recall_fscore
+from .metrics import hamming_distance
 from .metrics import mean_average_precision
+from .metrics import multi_hot_encode
+from .metrics import precision_recall_fscore
+from .misc import AverageMeter
+from .save_load import init_model, save_model

ppcls/utils/dist_utils.py

+# Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from typing import List, Union
+
+import paddle
+
+
+def all_gather(tensor: paddle.Tensor, concat: bool=True,
+               axis: int=0) -> Union[paddle.Tensor, List[paddle.Tensor]]:
+    """Gather tensor from all devices, concatenate them along given axis if specified.
+
+    Args:
+        tensor (paddle.Tensor): Tensor to be gathered from all GPUs.
+        concat (bool, optional): Whether to concatenate gathered Tensors. Defaults to True.
+        axis (int, optional): Axis which concatenated along. Defaults to 0.
+
+    Returns:
+        Union[paddle.Tensor, List[paddle.Tensor]]: Gathered Tensors
+    """
+    result = []
+    paddle.distributed.all_gather(result, tensor)
+    if concat:
+        return paddle.concat(result, axis)
+    return result